Thesis - Data, Word Embeddings, Summarization.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import time\n",
    "import dill\n",
    "from typing import List\n",
    "from functools import reduce\n",
    "from calendar import monthrange"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "import csv\n",
    "import requests\n",
    "import xml.etree.ElementTree as ET\n",
    "from bs4 import BeautifulSoup\n",
    "import urllib.request as urllib2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "from scipy import stats\n",
    "from random import sample, randint, shuffle\n",
    "from sklearn.decomposition import PCA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "import nltk\n",
    "import text_summarizer\n",
    "from gensim.utils import simple_preprocess\n",
    "from gensim.summarization.summarizer import summarize\n",
    "from gensim.models import Word2Vec, LsiModel\n",
    "from gensim.test.utils import common_dictionary, common_corpus"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from time import process_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "39.948508"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = process_time()\n",
    "dill.load_session('thesis_env.db')\n",
    "process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "393"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(adjectives)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1685787"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(all_)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Gender-neutral adjectives describing people"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('cities.csv') as csv_file:\n",
    "    cities = csv_file.read().split(\",\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "cities = [x.lower() for x in cities]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 373,
   "metadata": {},
   "outputs": [],
   "source": [
    "url = 'http://ideonomy.mit.edu/essays/traits.html'\n",
    "file = urllib2.urlopen(url)\n",
    "html = file.read()\n",
    "file.close()\n",
    "soup = BeautifulSoup(html)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 381,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Accessible'"
      ]
     },
     "execution_count": 381,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "soup.find_all('li')[0].contents[0].strip()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 389,
   "metadata": {},
   "outputs": [],
   "source": [
    "adjs = [x.contents[0].strip().lower() for x in soup.find_all('li')]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 386,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "393"
      ]
     },
     "execution_count": 386,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(adjectives)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 396,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['able',\n",
       " 'abnormal',\n",
       " 'above average',\n",
       " 'absent-minded',\n",
       " 'adventurous',\n",
       " 'affectionate',\n",
       " 'agile',\n",
       " 'agreeable',\n",
       " 'alert',\n",
       " 'amazing']"
      ]
     },
     "execution_count": 396,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "adjectives[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 399,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "882"
      ]
     },
     "execution_count": 399,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "adjs_all = adjs + [x for x in adjectives if x not in adjs]\n",
    "len(adjs_all)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 400,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "508"
      ]
     },
     "execution_count": 400,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len([x for x in adjs_all if x in vocab])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 397,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['able',\n",
       " 'abnormal',\n",
       " 'above average',\n",
       " 'absent-minded',\n",
       " 'affectionate',\n",
       " 'agile',\n",
       " 'amazing',\n",
       " 'analytical',\n",
       " 'angelic',\n",
       " 'apprehensive']"
      ]
     },
     "execution_count": 397,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "[x for x in adjectives if x not in adjs][:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 398,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['accessible',\n",
       " 'active',\n",
       " 'adaptable',\n",
       " 'admirable',\n",
       " 'adventurous',\n",
       " 'agreeable',\n",
       " 'alert',\n",
       " 'allocentric',\n",
       " 'amiable',\n",
       " 'anticipative']"
      ]
     },
     "execution_count": 398,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "adjs[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# adjs = [x.contents[0] for x in BeautifulSoup(html).find_all(attrs={\"class\": \"wordlist-item\"})]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [],
   "source": [
    "# url = 'https://www.enchantedlearning.com/wordlist/adjectivesforpeople.shtml'\n",
    "# file = urllib2.urlopen(url)\n",
    "# html = file.read()\n",
    "# file.close()\n",
    "# soup = BeautifulSoup(html)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# adjs = [x.contents[0] for x in BeautifulSoup(html).find_all(attrs={\"class\": \"wordlist-item\"})]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# with open('adjectives.csv', mode='w') as f:\n",
    "#     writer = csv.writer(f, delimiter=',')\n",
    "#     writer.writerow(adjs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('adjectives.csv') as csv_file:\n",
    "    adjectives = csv_file.read().split(\",\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['able', 'abnormal', 'above average', 'absent-minded', 'adventurous']"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "adjectives[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "dill.dump_session('thesis_env.db')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# url = 'https://www.britannica.com/topic/list-of-cities-and-towns-in-the-United-States-2023068'\n",
    "# file = urllib2.urlopen(url)\n",
    "# html = file.read()\n",
    "# file.close()\n",
    "# soup = BeautifulSoup(html)  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# alist = soup.find_all(attrs={\"class\": \"md-crosslink\"})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# places = [a.contents[0] for a in alist[3:]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2010"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# len(places)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "grammar = ['a', 'an', 'the', \\\n",
    "           'for', 'and', 'nor', 'but', 'or', 'yet', 'so', \\\n",
    "           'after', 'as', 'before', 'if', 'inasmuch', 'now', \\\n",
    "           'once', 'since', 'supposing', 'though', 'until', \\\n",
    "           'whenever', 'whereas', 'wherever', 'which', \\\n",
    "           'although', 'because', 'even', 'lest', 'both', 'either', \\\n",
    "           'hardly', 'when', 'neither', 'than', 'whether', \\\n",
    "           'consequently', 'finally', 'furthermore', 'hence', \\\n",
    "           'however', 'incidentally', 'indeed', 'instead', 'likewise', \\\n",
    "           'meanwhile', 'about', 'above', 'across', 'after', 'ago', \\\n",
    "           'at', 'below', 'by', 'during', 'for', 'in', \\\n",
    "           'on', 'over', 'past', 'since', 'through', 'to', \\\n",
    "           'up', 'amid', 'atop', 'onto', \\\n",
    "           'throughout', 'upon', 'within', 'without']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# with open('cities.csv', mode='w') as f:\n",
    "#     writer = csv.writer(f, delimiter=',')\n",
    "#     writer.writerow(places)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data Parsing and Processing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Doc:\n",
    "    def __init__(self, id: int, gdescriptors: List[str], descriptors: List[str], text: str) -> None:\n",
    "        self.id = id\n",
    "        self.gdescriptors = gdescriptors\n",
    "        self.descriptors = descriptors\n",
    "        self.text = text"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "def parse_doc(file: str) -> Doc:\n",
    "    tree = ET.parse(file)\n",
    "    root = tree.getroot()\n",
    "    try:\n",
    "        fulltext = root.find('body').find('body.content').find(\"*[@class='full_text']\")\n",
    "        text = reduce(lambda x, y: x + y, [x.text for x in fulltext.findall('p')])\n",
    "        id = int(root.find('head').find('docdata').find('doc-id').attrib['id-string'])\n",
    "        classifiers = root.find('head').find('docdata').find('identified-content')\n",
    "        gclassifiers = classifiers.findall(\"*[@class='online_producer'][@type='general_descriptor']\")\n",
    "        dclassifiers = classifiers.findall(\"*[@class='online_producer'][@type='descriptor']\")\n",
    "        gdescriptors = [c.text for c in gclassifiers]\n",
    "        descriptors = [c.text for c in dclassifiers]\n",
    "        return Doc(id, gdescriptors, descriptors, text)\n",
    "    except:\n",
    "        return None"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_docs_by_month(year: int, month: int) -> List[Doc]:\n",
    "    days = monthrange(year, month)[1]\n",
    "    docs = []\n",
    "    for day in range(1, days+1):\n",
    "        for file in os.listdir(f'data/{year}/{month:02}/{day:02}/'):\n",
    "            doc = parse_doc(f'data/{year}/{month:02}/{day:02}/{file}')\n",
    "            if doc:\n",
    "                docs.append(doc)\n",
    "    return docs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1988\n",
      "1989\n",
      "1990\n",
      "1991\n",
      "1992\n",
      "1993\n",
      "1994\n",
      "1995\n",
      "1996\n",
      "1997\n",
      "1998\n",
      "1999\n",
      "2000\n",
      "2001\n",
      "2002\n",
      "2003\n",
      "2004\n",
      "2005\n",
      "2006\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "973.445533"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "all_ = []\n",
    "for y in range(1988, 2007):\n",
    "    print(y)\n",
    "    for i in range(1,13):\n",
    "        all_ += get_docs_by_month(y, i)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1685787"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(all_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "dill.dump_session('thesis_env.db')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Saving files locally and loading"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('alldocs.json', 'w') as outfile:\n",
    "    json.dump([x.__dict__ for x in all_], outfile)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "def save(i, arr):\n",
    "    t = time.process_time()\n",
    "    with open('alldocs' + str(i) + '.json', 'w') as outfile:\n",
    "        json.dump([x.__dict__ for x in arr], outfile)\n",
    "    print(time.process_time() - t)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "def load(i):\n",
    "    t = time.process_time()\n",
    "    with open('alldocs' + str(i) + '.json', 'r') as infile:\n",
    "        l = json.load(infile)\n",
    "    print(time.process_time() - t)\n",
    "    return l"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "26.122622999999976\n"
     ]
    }
   ],
   "source": [
    "save(0, all_[:600000])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "89.333306\n"
     ]
    }
   ],
   "source": [
    "l0 = load(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "34.060252999999875\n"
     ]
    }
   ],
   "source": [
    "save(1, all_[600000:1200000])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "102.26382799999988\n"
     ]
    }
   ],
   "source": [
    "l1 = load(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "31.626627999999982\n"
     ]
    }
   ],
   "source": [
    "save(2, all_[1200000:])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "19.019525000000158\n"
     ]
    }
   ],
   "source": [
    "l2 = load(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1685787"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "l = l0 + l1 + l2\n",
    "len(l)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Process by descriptor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "44.409671"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "alld = {}\n",
    "for doc in all_:\n",
    "    doc = Doc(doc.id, doc.gdescriptors, doc.descriptors, doc.text)\n",
    "    for d in doc.descriptors:\n",
    "        count, docs = alld.get(d, (0, set()))\n",
    "        docs.add(doc)\n",
    "        alld[d] = (count + 1, docs)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('Terrorism', 12393)\n",
      "('Books and Literature', 11222)\n",
      "('Politics and Government', 10589)\n",
      "('Baseball', 9039)\n",
      "('United States Politics and Government', 8295)\n",
      "('Motion Pictures', 7748)\n",
      "('Music', 7317)\n",
      "('Weddings and Engagements', 7280)\n",
      "('Medicine and Health', 6829)\n",
      "('Travel and Vacations', 6537)\n",
      "('Deaths (Obituaries)', 6438)\n",
      "('Television', 6171)\n",
      "('Football', 6070)\n",
      "('Computers and the Internet', 5952)\n",
      "('Basketball', 5746)\n"
     ]
    }
   ],
   "source": [
    "alldlist = sorted([(d, alld[d][0], alld[d][1]) for d in alld], key=lambda x: x[1], reverse=True)\n",
    "for x in range(15):\n",
    "    print(alldlist[x][0:2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 141,
   "metadata": {},
   "outputs": [],
   "source": [
    "def exclude(e, x):\n",
    "    for d in x.descriptors:\n",
    "        if d == e:\n",
    "            continue\n",
    "        if d in [t[0] for t in alldlist[:30]]:\n",
    "            return False\n",
    "    return True"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def getlist(topic):\n",
    "    return list(filter(lambda x: exclude(topic, x), alld[topic][1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "metadata": {},
   "outputs": [],
   "source": [
    "politics = getlist('Politics and Government')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {},
   "outputs": [],
   "source": [
    "movies = getlist('Motion Pictures')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [],
   "source": [
    "books = getlist('Books and Literature')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 447,
   "metadata": {},
   "outputs": [],
   "source": [
    "sports = getlist('Baseball') + getlist('Football')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(9979, 7374, 10499, 7370)"
      ]
     },
     "execution_count": 112,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(politics), len(movies), len(books), len(sports)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "politics = sample(politics, 7370)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "movies = sample(movies, 7370)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "books = sample(books, 7370)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "sports = sample(sports, 7370)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# WordNet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 271,
   "metadata": {},
   "outputs": [],
   "source": [
    "from nltk.corpus import wordnet as wn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 281,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[Synset('person.n.01'), Synset('person.n.02'), Synset('person.n.03')]"
      ]
     },
     "execution_count": 281,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "wn.synsets('person')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 277,
   "metadata": {},
   "outputs": [],
   "source": [
    "person = wn.synset('person.n.01')\n",
    "man = wn.synset('man.n.01')\n",
    "woman = wn.synset('woman.n.01')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 279,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.3333333333333333, 0.3333333333333333, 0.3333333333333333)"
      ]
     },
     "execution_count": 279,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "man.path_similarity(person), woman.path_similarity(person), man.path_similarity(woman)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 352,
   "metadata": {},
   "outputs": [],
   "source": [
    "president = wn.synset('president.n.01')\n",
    "nurse = wn.synset('nurse.n.01')\n",
    "teacher = wn.synset('teacher.n.01')\n",
    "girl = wn.synset('girl.n.01')\n",
    "mother = wn.synset('mother.n.01')\n",
    "brother = wn.synset('brother.n.01')\n",
    "female = wn.synset('female.n.02')\n",
    "male = wn.synset('male.n.02')\n",
    "tall = wn.synset('tall.a.01')\n",
    "maid = wn.synset('maid.n.01')\n",
    "home = wn.synset('home.n.01')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 328,
   "metadata": {},
   "outputs": [],
   "source": [
    "smart.path_similarity(wn.synset('male.a.01'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 332,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.08333333333333333, 0.08333333333333333)"
      ]
     },
     "execution_count": 332,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "woman.path_similarity(home), man.path_similarity(home)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 333,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.1, 0.14285714285714285, 0.125, 0.125)"
      ]
     },
     "execution_count": 333,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mother.path_similarity(girl), mother.path_similarity(person), mother.path_similarity(female), mother.path_similarity(male)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 353,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.1111111111111111 0.1111111111111111 0.125 0.125\n",
      "0.2 0.2 0.16666666666666666 0.16666666666666666\n",
      "0.2 0.2 0.16666666666666666 0.16666666666666666\n",
      "0.25 0.5 0.2 0.3333333333333333\n",
      "0.1111111111111111 0.1111111111111111 0.125 0.125\n",
      "0.14285714285714285 0.14285714285714285 0.16666666666666666 0.16666666666666666\n",
      "0.16666666666666666 0.25 0.2 0.3333333333333333\n",
      "0.14285714285714285 0.14285714285714285 0.16666666666666666 0.16666666666666666\n",
      "0.25 0.5 0.3333333333333333 1.0\n"
     ]
    }
   ],
   "source": [
    "for x in [president, nurse, teacher, girl, mother, brother, schoolgirl, maid, female]:\n",
    "    print(man.path_similarity(x), woman.path_similarity(x), male.path_similarity(x), female.path_similarity(x))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SentiWordNet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 345,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[nltk_data] Downloading package sentiwordnet to\n",
      "[nltk_data]     /Users/mayjiang/nltk_data...\n",
      "[nltk_data]   Unzipping corpora/sentiwordnet.zip.\n"
     ]
    }
   ],
   "source": [
    "nltk.download('sentiwordnet')\n",
    "from nltk.corpus import sentiwordnet as swn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 371,
   "metadata": {},
   "outputs": [],
   "source": [
    "sman = swn.senti_synset('man.n.01')\n",
    "swoman = swn.senti_synset('woman.n.01')\n",
    "tall = swn.senti_synset('tall.a.01')\n",
    "maid = swn.senti_synset('maid.n.01')\n",
    "smart = swn.senti_synset('smart.a.01')\n",
    "awesome = swn.senti_synset('awesome.a.01')\n",
    "shy = swn.senti_synset('shy.a.01')\n",
    "liberal = swn.senti_synset('liberal.a.01')\n",
    "conservative = swn.senti_synset('conservative.a.01')\n",
    "moderate = swn.senti_synset('moderate.a.01')\n",
    "demanding = swn.senti_synset('demanding.a.01')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 372,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<man.n.01: PosScore=0.0 NegScore=0.0>\n",
      "<woman.n.01: PosScore=0.0 NegScore=0.0>\n",
      "<tall.a.01: PosScore=0.5 NegScore=0.0>\n",
      "<maid.n.01: PosScore=0.0 NegScore=0.0>\n",
      "<smart.a.01: PosScore=0.5 NegScore=0.0>\n",
      "<amazing.s.02: PosScore=0.875 NegScore=0.125>\n",
      "<diffident.a.02: PosScore=0.25 NegScore=0.25>\n",
      "<broad.s.08: PosScore=0.625 NegScore=0.0>\n",
      "<conservative.a.01: PosScore=0.0 NegScore=0.0>\n",
      "<moderate.a.01: PosScore=0.25 NegScore=0.625>\n",
      "<demanding.a.01: PosScore=0.0 NegScore=0.375>\n"
     ]
    }
   ],
   "source": [
    "for s in [sman, swoman, tall, maid, smart, awesome, shy, liberal, conservative, moderate, demanding]:\n",
    "    print(s)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 401,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Sentiment Bias - weighted projection bias, each word multiplied by (pos_score - neg_score)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 441,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "neg -0.20761971 0.25 caring\n",
      "pos 0.12445271 0.375 confident\n",
      "pos 0.10769328 0.25 contemplative\n",
      "pos 0.16988644 0.25 firm\n",
      "neg -0.16727082 0.125 friendly\n",
      "pos 0.17441696 0.625 liberal\n",
      "pos 0.13595861 0.625 logical\n",
      "pos 0.27562273 -0.375 moderate\n",
      "neg -0.17365715 0.25 patient\n",
      "pos 0.14351737 0.125 patriotic\n",
      "pos 0.1057783 0.375 rational\n",
      "neg -0.25844532 0.625 sexy\n",
      "neg -0.18216969 0.375 warm\n",
      "pos 0.11109869 0.5 winning\n",
      "neg -0.21810304 0.5 cute\n",
      "neg -0.22588326 0.375 emotional\n",
      "neg -0.21726993 0.375 glamorous\n",
      "pos 0.11444926 0.625 moralistic\n",
      "pos 0.10260116 0.625 progressive\n",
      "neg -0.25614166 0.125 sensual\n",
      "neg -0.17997913 0.125 anxious\n",
      "pos 0.104796335 -0.75 cold\n",
      "pos 0.23928382 -0.875 criminal\n",
      "pos 0.11037847 -0.5 critical\n",
      "neg -0.15172711 -0.625 discontented\n",
      "neg -0.15982822 0.125 disloyal\n",
      "pos 0.17313033 -0.125 fanatical\n",
      "pos 0.26991513 0.25 grand\n",
      "pos 0.10619152 0.125 hidebound\n",
      "pos 0.15393595 -0.375 moody\n",
      "pos 0.19790758 -0.75 paranoid\n",
      "pos 0.10387065 0.375 provocative\n",
      "pos 0.15288518 -0.5 reactionary\n",
      "neg -0.16902468 0.125 rowdy\n",
      "neg -0.17762111 -0.5 stiff\n",
      "neg -0.15178116 0.75 submissive\n",
      "pos 0.10755275 -0.875 troublesome\n",
      "neg -0.15146762 -0.75 unrealistic\n",
      "pos 0.12196509 0.625 venal\n",
      "neg -0.31382367 0.75 beautiful\n",
      "neg -0.17306766 0.5 chic\n",
      "pos 0.13684565 0.125 civil\n",
      "neg -0.16664672 0.5 demure\n",
      "neg -0.30375844 0.75 fabulous\n",
      "pos 0.10945127 0.5 indefatigable\n",
      "pos 0.10757194 0.625 jovial\n",
      "neg -0.21401869 0.625 lovely\n",
      "neg -0.19062826 -0.625 plucky\n",
      "neg -0.15392375 0.75 pretty\n",
      "pos 0.13982296 0.875 respected\n",
      "pos 0.100318715 0.5 revered\n",
      "neg -0.20907472 -0.125 sassy\n",
      "neg -0.16599339 0.5 smart\n",
      "neg -0.16473676 -0.25 testy\n",
      "pos 0.10595325 -0.75 unthinking\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.061671779339609586, 0.08620689655172414, 0.21634615384615385)"
      ]
     },
     "execution_count": 441,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sbias(getg(mbooks), mbooks)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 440,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "neg -0.19597507 0.625 accessible\n",
      "pos 0.15286292 0.5 charismatic\n",
      "pos 0.11145699 0.125 conciliatory\n",
      "neg -0.15788443 0.875 decent\n",
      "neg -0.16832036 0.75 healthy\n",
      "neg -0.16127829 0.25 intelligent\n",
      "pos 0.19596937 0.5 loyal\n",
      "neg -0.19834132 0.125 neat\n",
      "neg -0.35829028 0.25 patient\n",
      "neg -0.17002498 0.75 precise\n",
      "neg -0.16689934 -0.25 simple\n",
      "pos 0.11683028 0.25 steadfast\n",
      "pos 0.13272035 0.625 suave\n",
      "pos 0.10439743 0.5 winning\n",
      "neg -0.16523525 0.375 busy\n",
      "pos 0.19985822 -0.75 earthy\n",
      "neg -0.15106884 -0.125 ordinary\n",
      "pos 0.12091027 -0.125 outspoken\n",
      "neg -0.17384371 -0.125 private\n",
      "neg -0.16532537 0.25 pure\n",
      "pos 0.11442726 0.25 smooth\n",
      "pos 0.14168674 0.5 crafty\n",
      "pos 0.12447191 0.125 disloyal\n",
      "neg -0.16128409 -0.625 disobedient\n",
      "pos 0.12090054 -0.25 disorganized\n",
      "pos 0.12598191 0.375 fatalistic\n",
      "pos 0.112024866 -0.125 gloomy\n",
      "pos 0.18825054 -0.375 mannered\n",
      "pos 0.106990054 0.125 ruined\n",
      "neg -0.17413384 0.25 amazing\n",
      "neg -0.20340455 0.75 beautiful\n",
      "neg -0.16406856 0.125 bright\n",
      "pos 0.103838444 -0.625 caustic\n",
      "pos 0.10076103 0.125 frank\n",
      "pos 0.2035518 -0.25 lame\n",
      "neg -0.17072235 -0.875 nasty\n",
      "neg -0.21208668 -0.5 normal\n",
      "neg -0.15433383 0.625 positive\n",
      "pos 0.13305448 -0.125 reclusive\n",
      "pos 0.12587817 0.5 revered\n",
      "pos 0.10030437 0.625 sincere\n",
      "pos 0.10856759 -0.625 sneering\n",
      "pos 0.16759476 -0.25 somber\n",
      "pos 0.10130627 -0.125 unhelpful\n",
      "neg -0.16317654 -0.5 unpleasant\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.06415719592100227, 0.06, 0.15)"
      ]
     },
     "execution_count": 440,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sbias(getg(mpol), mpol)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 420,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.24627662, -0.45324638)"
      ]
     },
     "execution_count": 420,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gproj(getg(mpol), mpol, 'he'), gproj(getg(mpol), mpol, 'she')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 439,
   "metadata": {},
   "outputs": [],
   "source": [
    "def sbias(g, m):\n",
    "    b = 0\n",
    "    l = list(filter(lambda x: x in m.wv.vocab, adjs_all))\n",
    "    p, n, np, nn = 0, 0, 0, 0\n",
    "    for x in l:\n",
    "        proj = gproj(g, m, x)\n",
    "        b += abs(proj)\n",
    "        try:\n",
    "            s = swn.senti_synset(x + \".a.01\")\n",
    "            score = s.pos_score() - s.neg_score()\n",
    "            if proj < -0.15 and score != 0:\n",
    "                print(\"neg\", proj, score, x)\n",
    "                n += score\n",
    "                nn += 1\n",
    "            if proj > 0.1 and score != 0:\n",
    "                print(\"pos\", proj, score, x)\n",
    "                p += score\n",
    "                np += 1\n",
    "        except:\n",
    "            continue\n",
    "    return b/len(l), p/np, n/nn"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Experiment: Male/Female Ratio"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 180,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Terrorism 6899\n",
      "Books and Literature 7787\n",
      "Politics and Government 5649\n",
      "Baseball 8270\n",
      "United States Politics and Government 3634\n",
      "Motion Pictures 5895\n",
      "Music 4838\n",
      "Weddings and Engagements 7167\n",
      "Medicine and Health 5079\n",
      "Travel and Vacations 4555\n",
      "Deaths (Obituaries) 3738\n",
      "Television 4294\n",
      "Football 5593\n",
      "Computers and the Internet 4201\n",
      "Basketball 5410\n",
      "Art 3846\n",
      "Theater 3901\n",
      "Suits and Litigation 3275\n",
      "United States International Relations 2496\n",
      "United States Armament and Defense 2212\n",
      "International Relations 2631\n",
      "Children and Youth 2207\n",
      "Education and Schools 3074\n",
      "Elections 2068\n",
      "Environment 2415\n",
      "Housing 2784\n",
      "Automobiles 2639\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "27"
      ]
     },
     "execution_count": 180,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lists = []\n",
    "labels = []\n",
    "for c in alldlist[:30]:\n",
    "    l = getlist(c[0])\n",
    "    if len(l) > 2000:\n",
    "        print(c[0], len(l))\n",
    "        lists.append(l)\n",
    "        labels.append(c[0])\n",
    "len(lists)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 146,
   "metadata": {},
   "outputs": [],
   "source": [
    "def heshestats(m, w1, w2):\n",
    "    c1 = m.wv.vocab[w1].count #+ m.wv.vocab[w1_].count\n",
    "    c2 = m.wv.vocab[w2].count #+ m.wv.vocab[w2_].count\n",
    "    #print(c1, c2, c2/c1)\n",
    "    return c2/c1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "14443 2707 0.18742643495118744\n",
      "18728 8495 0.4535988893635199\n",
      "16770 2663 0.15879546809779368\n",
      "29126 492 0.016892123875575087\n",
      "17186 2469 0.14366344699173747\n",
      "15933 6101 0.3829159605849495\n",
      "14728 5602 0.38036393264530144\n",
      "5119 5182 1.0123070912287555\n",
      "12091 6448 0.5332892233892979\n",
      "6763 3332 0.49268076297501107\n",
      "23219 4969 0.21400577113570782\n",
      "14508 5507 0.37958367797077475\n",
      "23644 580 0.024530536288276096\n",
      "10146 3115 0.30701754385964913\n",
      "22276 1743 0.07824564553779853\n",
      "16604 5778 0.34798843652132017\n",
      "13167 7451 0.5658844079896711\n",
      "13943 4111 0.2948432905400559\n",
      "12174 1353 0.11113849186791523\n",
      "14805 2654 0.17926376224248564\n",
      "10496 1374 0.13090701219512196\n",
      "14065 11199 0.7962317810167081\n",
      "11147 6672 0.5985466941778057\n",
      "18823 3508 0.18636774159273228\n",
      "8298 2153 0.2594601108700892\n",
      "12787 6793 0.5312426683350278\n",
      "10026 1905 0.19000598444045483\n"
     ]
    }
   ],
   "source": [
    "ms = []\n",
    "for l in lists:\n",
    "    m = getw2v(sample(l, 2000))\n",
    "    heshestats(m, 'he', 'she')\n",
    "    ms.append(m)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 147,
   "metadata": {},
   "outputs": [],
   "source": [
    "heshes = []\n",
    "for m in ms:\n",
    "    heshes.append(heshestats(m, 'he', 'she'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 181,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.18742643495118744 0.10755114227638267 Terrorism\n",
      "0.4535988893635199 0.04351092342586462 Books and Literature\n",
      "0.15879546809779368 0.08117492493491607 Politics and Government\n",
      "0.016892123875575087 0.23906051099345269 Baseball\n",
      "0.14366344699173747 0.10965769848256052 United States Politics and Government\n",
      "0.3829159605849495 0.07131090098361408 Motion Pictures\n",
      "0.38036393264530144 0.0511600704294268 Music\n",
      "1.0123070912287555 0.1493787405409173 Weddings and Engagements\n",
      "0.5332892233892979 0.0746384320254947 Medicine and Health\n",
      "0.49268076297501107 0.039487039247005425 Travel and Vacations\n",
      "0.21400577113570782 0.0834943001660846 Deaths (Obituaries)\n",
      "0.37958367797077475 0.06178147751563449 Television\n",
      "0.024530536288276096 0.23035889131755666 Football\n",
      "0.30701754385964913 0.11230526598537284 Computers and the Internet\n",
      "0.07824564553779853 0.15911339820761766 Basketball\n",
      "0.34798843652132017 0.03524149366149776 Art\n",
      "0.5658844079896711 0.056935037403313 Theater\n",
      "0.2948432905400559 0.06072088706943589 Suits and Litigation\n",
      "0.11113849186791523 0.08476496551964441 United States International Relations\n",
      "0.17926376224248564 0.10880618581677481 United States Armament and Defense\n",
      "0.13090701219512196 0.10391273738562383 International Relations\n",
      "0.7962317810167081 0.06930799046182697 Children and Youth\n",
      "0.5985466941778057 0.06576247124122718 Education and Schools\n",
      "0.18636774159273228 0.05862010479495102 Elections\n",
      "0.2594601108700892 0.12195312360464595 Environment\n",
      "0.5312426683350278 0.047254450624937556 Housing\n",
      "0.19000598444045483 0.3440743385858479 Automobiles\n"
     ]
    }
   ],
   "source": [
    "biases = []\n",
    "i = 0\n",
    "for m in ms:\n",
    "    b = bias(getg(m), m, m.wv.vocab, proflist)\n",
    "    print(heshestats(m, 'he', 'she'), b, labels[i])\n",
    "    i += 1\n",
    "    biases.append(b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 182,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "26"
      ]
     },
     "execution_count": 182,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels.index(\"Automobiles\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 215,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def coords(label):\n",
    "    i = labels.index(label)\n",
    "    return (heshes[i], biases[i])\n",
    "plt.plot(heshes, biases, \"o\")\n",
    "plt.title(\"Ratio of 'she' to 'he' vs Projection Bias on Professions\")\n",
    "plt.annotate(\"Automobiles\", coords(\"Automobiles\"), textcoords=\"offset points\", xytext=(5,0))\n",
    "plt.annotate(\"Weddings\", coords(\"Weddings and Engagements\"), textcoords=\"offset points\", xytext=(-52,0))\n",
    "plt.annotate(\"Baseball\", coords(\"Baseball\"), textcoords=\"offset points\", xytext=(5,0))\n",
    "plt.annotate(\"Football\", coords(\"Football\"), textcoords=\"offset points\", xytext=(5,-5))\n",
    "plt.annotate(\"Basketball\", coords(\"Basketball\"), textcoords=\"offset points\", xytext=(0,5))\n",
    "plt.annotate(\"Children\", coords(\"Children and Youth\"), textcoords=\"offset points\", xytext=(5,-7))\n",
    "plt.annotate(\"Education\", coords(\"Education and Schools\"), textcoords=\"offset points\", xytext=(2,5))\n",
    "plt.annotate(\"Computers\", coords(\"Computers and the Internet\"), textcoords=\"offset points\", xytext=(5,0))\n",
    "plt.annotate(\"Politics\", coords(\"Politics and Government\"), textcoords=\"offset points\", xytext=(22,0))\n",
    "plt.annotate(\"Travel\", coords(\"Travel and Vacations\"), textcoords=\"offset points\", xytext=(2,-10))\n",
    "plt.annotate(\"Art\", coords(\"Art\"), textcoords=\"offset points\", xytext=(-18,-5))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Experiment: Length"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# books is 10499 long, split into increments of 500"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [],
   "source": [
    "bookslist = []\n",
    "for i in range(1, 21):\n",
    "    bookslist.append(sample(books, i*500))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {},
   "outputs": [],
   "source": [
    "plist = []\n",
    "for i in range(2, 20):\n",
    "    plist.append(sample(politics, i*500))\n",
    "plist.append(politics)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 221,
   "metadata": {},
   "outputs": [],
   "source": [
    "plistshort = []\n",
    "for i in range(1, 10):\n",
    "    plistshort.append(sample(politics, i*200))\n",
    "pmodelsshort = [getw2v(p) for p in plistshort]\n",
    "pbiasshort = [bias(getg(m), m, m.wv.vocab, proflist) for m in pmodelsshort]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 224,
   "metadata": {},
   "outputs": [],
   "source": [
    "blistshort = []\n",
    "for i in range(1, 10):\n",
    "    blistshort.append(sample(books, i*200))\n",
    "bmodelsshort = [getw2v(b) for b in blistshort]\n",
    "bbiasshort = [bias(getg(m), m, m.wv.vocab, proflist) for m in bmodelsshort]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 448,
   "metadata": {},
   "outputs": [],
   "source": [
    "slistshort = []\n",
    "for i in range(1, 10):\n",
    "    slistshort.append(sample(sports, i*200))\n",
    "smodelsshort = [getw2v(s) for s in slistshort]\n",
    "sbiasshort = [bias(getg(m), m, m.wv.vocab, proflist) for m in smodelsshort]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 451,
   "metadata": {},
   "outputs": [],
   "source": [
    "slist = []\n",
    "for i in range(2, 14):\n",
    "    slist.append(sample(sports, i*1000))\n",
    "smodels = [getw2v(s) for s in slist]\n",
    "sbias = [bias(getg(m), m, m.wv.vocab, proflist) for m in smodels]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 449,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot([i*200 for i in range(1,10)], bbiasshort, label=\"books\")\n",
    "plt.plot([i*200 for i in range(1,10)], pbiasshort, label=\"politics\")\n",
    "plt.legend()\n",
    "plt.title(\"Projection Bias on Professions vs Number of Articles \\n in Training Set (200-1800 articles)\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {},
   "outputs": [],
   "source": [
    "pmodels = [getw2v(p) for p in plist]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [],
   "source": [
    "booksmodels = [getw2v(b) for b in bookslist]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.5772233574508546\n",
      "0.0734701767039027\n",
      "0.0688537814152118\n",
      "0.08305388300997958\n",
      "0.06266421009538382\n",
      "0.05534665679485859\n",
      "0.07938607440166894\n",
      "0.06768135385776182\n",
      "0.08162163002018975\n",
      "0.08094774881021471\n",
      "0.07494355367683826\n",
      "0.09051752739804284\n",
      "0.09702742212223557\n",
      "0.10574751904139042\n",
      "0.10613597846505556\n",
      "0.10079952142610644\n",
      "0.09749374895855079\n",
      "0.09925294766243084\n",
      "0.10362135741110567\n"
     ]
    }
   ],
   "source": [
    "pbias = []\n",
    "for m in pmodels:\n",
    "    b = bias(getg(m), m, m.wv.vocab, proflist)\n",
    "    print(b)\n",
    "    pbias.append(b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9727325638135275\n",
      "0.045612330736406935\n",
      "0.04454958280516756\n",
      "0.04608471256797202\n",
      "0.05587483647625891\n",
      "0.05342771026891355\n",
      "0.06256595694921771\n",
      "0.0673831163421047\n",
      "0.06619414358791284\n",
      "0.06855078208813771\n",
      "0.075337888015471\n",
      "0.07627692876604004\n",
      "0.08308898592561793\n",
      "0.08104289494451168\n",
      "0.08403255242510363\n",
      "0.08511889418915994\n",
      "0.09022001865134682\n",
      "0.09121858018543917\n",
      "0.09542696736488897\n",
      "0.10060866628945977\n"
     ]
    }
   ],
   "source": [
    "bookbias = []\n",
    "for bm in booksmodels:\n",
    "    b = bias(getg(bm), bm, bm.wv.vocab, proflist)\n",
    "    print(b)\n",
    "    bookbias.append(b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 453,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot([i*500 for i in range(3,21)], bookbias[2:], label=\"books\")\n",
    "plt.plot([i*500 for i in range(3,21)], pbias[1:], label=\"politics\")\n",
    "plt.legend()\n",
    "plt.title(\"Projection Bias on Professions vs Number of Articles \\n in Training Set (1500-10000 articles)\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 458,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.19302710774604992,\n",
       " 0.23844998456397146,\n",
       " 0.2893524986475313,\n",
       " 0.27873200980463514,\n",
       " 0.2543005048512266,\n",
       " 0.3116570048239882,\n",
       " 0.3102601092703866,\n",
       " 0.30372024667422987,\n",
       " 0.305594824975516,\n",
       " 0.3274015368861792,\n",
       " 0.35638563316315414,\n",
       " 0.34060144879534776]"
      ]
     },
     "execution_count": 458,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot([i*1000 for i in range(2,14)], sbias, label=\"s\")\n",
    "sbias"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 442,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot([i*200 for i in range(1,10)] + [i*500 for i in range(3,21)], bbiasshort + bookbias[2:], label=\"books\")\n",
    "plt.plot([i*200 for i in range(1,10)] + [i*500 for i in range(3,21)], pbiasshort + pbias[1:], label=\"politics\")\n",
    "plt.legend()\n",
    "plt.title(\"Projection Bias on Professions vs Number of Articles \\n in Training Set (200-1800 articles)\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 267,
   "metadata": {},
   "outputs": [],
   "source": [
    "stereotypes_m = ['captain', 'president', 'principal', 'dean', \\\n",
    "                 'coach', 'director', 'officer', 'manager']\n",
    "stereotypes_f = ['teacher', 'author', 'secretary', 'writer', \\\n",
    "                 'singer', 'teenager', 'editor', 'photographer']\n",
    "male = [\"male\", \"man\", \"boy\", \"boys\", \"brother\", \"he\", \"him\", \"son\", \"father\"]\n",
    "female = [\"female\", \"woman\", \"girl\", \"girls\", \"sister\", \"she\", \"her\", \\\n",
    "          \"daughter\", \"mother\"]\n",
    "def cos_sim(m, x, y):\n",
    "    wv1 = m.wv.get_vector(x)\n",
    "    wv2 = m.wv.get_vector(y)\n",
    "    sim = np.dot(wv1, wv2)/(np.linalg.norm(wv1)*np.linalg.norm(wv2))\n",
    "    return sim\n",
    "def s_word(m, w, A, B):\n",
    "    return np.mean([cos_sim(m, w, a) for a in A]) - np.mean([cos_sim(m, w, b) for b in B])\n",
    "def effect_size(m, X, Y, A, B):\n",
    "    num = np.mean([s_word(m, x, A, B) for x in X]) - np.mean([s_word(m, y, A, B) for y in Y])\n",
    "    return num/np.std([s_word(m, x, A, B) for x in X + Y])\n",
    "def wbias(m):\n",
    "    return effect_size(m, male, female, stereotypes_m, stereotypes_f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 268,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.084710464\n",
      "-0.103254676\n",
      "0.46582845\n",
      "0.17834492\n",
      "0.92044437\n",
      "0.99255073\n",
      "1.0072471\n",
      "1.0636849\n",
      "0.95142347\n",
      "1.0167873\n",
      "1.0270476\n",
      "1.1150552\n",
      "1.0032507\n",
      "1.0011431\n",
      "1.2693148\n",
      "1.27527\n",
      "1.2872465\n",
      "1.2055503\n",
      "1.1214428\n",
      "1.2525442\n"
     ]
    }
   ],
   "source": [
    "bookwbias = []\n",
    "for bm in booksmodels:\n",
    "    b = wbias(bm)\n",
    "    print(b)\n",
    "    bookwbias.append(b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 269,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.53966504\n",
      "0.58441985\n",
      "0.6321938\n",
      "0.4419999\n",
      "0.52009445\n",
      "0.40268213\n",
      "0.45781055\n",
      "0.25239307\n",
      "0.45995572\n",
      "0.27195734\n",
      "0.9551928\n",
      "0.6784737\n",
      "0.39555675\n",
      "0.49233136\n",
      "0.64562297\n",
      "0.8364172\n",
      "0.5080821\n",
      "0.72874755\n",
      "0.53683764\n"
     ]
    }
   ],
   "source": [
    "pwbias = []\n",
    "for m in pmodels:\n",
    "    b = wbias(m)\n",
    "    print(b)\n",
    "    pwbias.append(b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 270,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot([i*500 for i in range(3,21)], bookwbias[2:], label=\"books\")\n",
    "plt.plot([i*500 for i in range(3,21)], pwbias[1:], label=\"politics\")\n",
    "plt.legend()\n",
    "plt.title(\"WEAT Bias on Professions vs Number of Articles \\n in Training Set (1500-10000 articles)\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training Word Embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [],
   "source": [
    "def getw2v(category):\n",
    "    sentences = []\n",
    "    for t in category:\n",
    "        sentences += nltk.sent_tokenize(t.text)\n",
    "    docs = [simple_preprocess(s) for s in sentences]\n",
    "    return Word2Vec(docs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "95.68879"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "msports = getw2v(sports)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "163.753283"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "msports1 = getw2v(sports + sports)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "371.57096499999994"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "msports2 = getw2v(sports + sports + sports + sports)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "141.71859400000005"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "mpol = getw2v(politics)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "187.84100700000022"
      ]
     },
     "execution_count": 68,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "mpol1 = getw2v(politics + politics)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "352.6542690000001"
      ]
     },
     "execution_count": 69,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "mpol2 = getw2v(politics + politics + politics + politics)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "115.31121599999994"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "mmov = getw2v(movies)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "261.3358850000004"
      ]
     },
     "execution_count": 70,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "mmov1 = getw2v(movies + movies)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "421.18423699999994"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "mmov2 = getw2v(movies + movies + movies + movies)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "118.71266199999991"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "mbooks = getw2v(books)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "231.03794900000003"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "mbooks1 = getw2v(books + books)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "499.228335"
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "mbooks2 = getw2v(books + books + books + books)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def top5(word):\n",
    "    for m in [mpol, mbooks, mmov, msports]:\n",
    "        print([x[0] for x in m.wv.most_similar(positive=[word], topn=5)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 198,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['supremacist', 'collar', 'plains', 'assembly', 'congressional']\n",
      "['family', 'apartment', 'petal', 'mansion', 'farmhouse']\n",
      "['mansion', 'houses', 'neighborhood', 'castle', 'cemetery']\n",
      "['apartment', 'restaurant', 'store', 'shop', 'queens']\n"
     ]
    }
   ],
   "source": [
    "top5('house')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 199,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['difficulty', 'credibility', 'weakness', 'vision', 'strength']\n",
      "['popularity', 'achievement', 'goal', 'fame', 'criticism']\n",
      "['popularity', 'ambitions', 'hit', 'decline', 'commercial']\n",
      "['experience', 'talent', 'confidence', 'patience', 'swagger']\n"
     ]
    }
   ],
   "source": [
    "top5('success')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 200,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['seats', 'district', 'race', 'nomination', 'contest']\n",
      "['chair', 'plane', 'truck', 'door', 'bedroom']\n",
      "['chair', 'window', 'bed', 'truck', 'kitchen']\n",
      "['seats', 'suite', 'apartment', 'gate', 'windows']\n"
     ]
    }
   ],
   "source": [
    "top5('seat')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 201,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['good', 'greatest', 'whatever', 'whole', 'great']\n",
      "['greatest', 'top', 'nonfiction', 'list', 'finest']\n",
      "['oscars', 'prize', 'oscar', 'award', 'nominated']\n",
      "['worst', 'greatest', 'biggest', 'strongest', 'toughest']\n"
     ]
    }
   ],
   "source": [
    "top5(\"best\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Summarization"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## TextRank"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "def getsummaries(category):\n",
    "    s = []\n",
    "    fails = []\n",
    "    for i in range(len(category)):\n",
    "        try:\n",
    "            s.append(summarize(category[i].text, word_count=200))\n",
    "        except:\n",
    "            fails.append(i)\n",
    "            print(i)\n",
    "    return s, fails"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "t = time.process_time()\n",
    "summaries_pol, fpol = getsummaries(politics)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "166\n",
      "1831\n",
      "5320\n",
      "5643\n",
      "6149\n",
      "7366\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "109.888758"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "summaries_pol, fpol = getsummaries(politics)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1591\n",
      "3842\n",
      "6966\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "122.95149700000002"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "summaries_mov, fmov = getsummaries(movies)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "238\n",
      "5562\n",
      "6553\n",
      "7021\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "121.83292399999993"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "summaries_bks, fbks = getsummaries(books)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Centroid BOW Summarizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [],
   "source": [
    "def getbows(category):\n",
    "    s = []\n",
    "    centroid_summarizer = text_summarizer.CentroidBOWSummarizer()\n",
    "    for i in category:\n",
    "        try:\n",
    "            s.append(centroid_summarizer.summarize(i.text, limit=200-30))\n",
    "        except:\n",
    "            print(i)\n",
    "    return s"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "134.52433099999996"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "bow_pol = getbows(politics)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "7370"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(bow_pol)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "147.18170899999996"
      ]
     },
     "execution_count": 83,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "bow_bks = getbows(books)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "135.9884790000001"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "bow_mov = getbows(movies)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Pseudo-summaries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "def getctrls(category):\n",
    "    ctrls = []\n",
    "    for a in category:\n",
    "        sents = nltk.sent_tokenize(a.text)\n",
    "        rs = [i for i in range(len(sents))]\n",
    "        shuffle(rs)\n",
    "        ctrl = []\n",
    "        count = 0\n",
    "        for r in rs:\n",
    "            if count > 200-5:\n",
    "                break\n",
    "            count += len(nltk.word_tokenize(sents[r]))\n",
    "            ctrl.append(sents[r])\n",
    "        summary = \"\\n\".join([s for s in ctrl])\n",
    "        ctrls.append(summary)\n",
    "    return ctrls"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "21.034067999999934"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "ctrls_pol = getctrls(politics)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "22.829864999999927"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "ctrls_bks = getctrls(books)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "22.42926299999999"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t = time.process_time()\n",
    "ctrls_mov = getctrls(movies)\n",
    "time.process_time() - t"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training Word Embeddings on Summaries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "def getw2v_s(sums):\n",
    "    sentences = []\n",
    "    for t in sums:\n",
    "        sentences += nltk.sent_tokenize(t)\n",
    "    docs = [simple_preprocess(s) for s in sentences]\n",
    "    return Word2Vec(docs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "mpol_s = getw2v_s(summaries_pol)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "mbks_s = getw2v_s(summaries_bks)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "mmov_s = getw2v_s(summaries_mov)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [],
   "source": [
    "# mpol_b = getw2v_s(bow_pol)\n",
    "mbks_b = getw2v_s(bow_bks)\n",
    "# mmov_b = getw2v_s(bow_mov)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "mpol_c = getw2v_s(ctrls_pol)\n",
    "mbks_c = getw2v_s(ctrls_bks)\n",
    "mmov_c = getw2v_s(ctrls_mov)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 163,
   "metadata": {},
   "outputs": [],
   "source": [
    "dill.dump_session('thesis_env2.db')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data Count"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 164,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(7370, 7370, 7370, 7370)"
      ]
     },
     "execution_count": 164,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(politics), len(books), len(movies), len(sports)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 204,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1022.20434192673\n",
      "1204.5089552238805\n",
      "1174.7398914518317\n",
      "975.4926729986431\n"
     ]
    }
   ],
   "source": [
    "for x in [politics, books, movies, sports]:\n",
    "    print(np.mean([len(nltk.word_tokenize(a.text)) for a in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 167,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "7533646\n",
      "8877231\n",
      "8657833\n",
      "7189381\n"
     ]
    }
   ],
   "source": [
    "for x in [politics, books, movies, sports]:\n",
    "    print(sum([len(nltk.word_tokenize(a.text)) for a in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 169,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1651028\n",
      "1728963\n",
      "1776150\n",
      "1693673\n"
     ]
    }
   ],
   "source": [
    "for x in [summaries_pol, summaries_bks, summaries_mov, summaries_spts]:\n",
    "    print(sum([len(nltk.word_tokenize(a)) for a in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 170,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1654079\n",
      "1664973\n",
      "1686927\n",
      "1594135\n"
     ]
    }
   ],
   "source": [
    "for x in [bow_pol, bow_bks, bow_mov, bow_spts]:\n",
    "    print(sum([len(nltk.word_tokenize(a)) for a in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 175,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1651997\n",
      "1610595\n",
      "1624683\n",
      "1594895\n"
     ]
    }
   ],
   "source": [
    "for x in [ctrls_pol, ctrls_bks, ctrls_mov, ctrls_spts]:\n",
    "    print(sum([len(nltk.word_tokenize(a)) for a in x]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for m in [mpol, mbooks, mmov, msports]:\n",
    "    print(m.wv.vocab['she'].count, m.wv.vocab['he'].count)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 183,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "7827 8938\n",
      "36723 12895\n",
      "26108 10990\n",
      "1891 20541\n"
     ]
    }
   ],
   "source": [
    "for m in [mpol, mbooks, mmov, msports]:\n",
    "    print(m.wv.vocab['her'].count, m.wv.vocab['him'].count)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 184,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "990 2431\n",
      "3629 7280\n",
      "3336 6874\n",
      "191 2440\n"
     ]
    }
   ],
   "source": [
    "for m in [mpol, mbooks, mmov, msports]:\n",
    "    print(m.wv.vocab['woman'].count, m.wv.vocab['man'].count)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Bias"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.35275307\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.32663773588579276"
      ]
     },
     "execution_count": 144,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "vocab = vocab & set(msports.wv.vocab)\n",
    "gspts = getg(msports)\n",
    "bias(gspts, msports, vocab, proflist)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 146,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "295\n",
      "4071\n",
      "4800\n",
      "5414\n",
      "7092\n"
     ]
    }
   ],
   "source": [
    "summaries_spts, fspts = getsummaries(sports)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 147,
   "metadata": {},
   "outputs": [],
   "source": [
    "mspts_s = getw2v_s(summaries_spts)\n",
    "bow_spts = getbows(sports)\n",
    "mspts_b = getw2v_s(bow_spts)\n",
    "ctrls_spts = getctrls(sports)\n",
    "mspts_c = getw2v_s(ctrls_spts)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 145,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<gensim.models.word2vec.Word2Vec at 0x185ff4400>"
      ]
     },
     "execution_count": 145,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# mpol, mmov, mbooks, msports\n",
    "# summaries_{pol, mov, bks, spts}\n",
    "# m{pol, mov, bks}_{s, b, c}\n",
    "# bow_{pol, mov, bks}\n",
    "# ctrls_{pol, mov, bks}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 156,
   "metadata": {},
   "outputs": [],
   "source": [
    "mdict = {\"politics\": [], \"books\": [], \"movies\": [], \"sports\": []}\n",
    "mdictadd([mpol, mmov, mbooks, msports])\n",
    "mdictadd([mpol_s, mbks_s, mmov_s, mspts_s])\n",
    "mdictadd([mpol_b, mbks_b, mmov_b, mspts_b])\n",
    "mdictadd([mpol_c, mbks_c, mmov_c, mspts_c])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 157,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.34343678\n",
      "0.52675015\n",
      "0.5915069\n",
      "0.35275307\n",
      "0.31522945\n",
      "0.78374213\n",
      "0.62817353\n",
      "0.49360958\n",
      "0.31965494\n",
      "0.8359838\n",
      "0.7387123\n",
      "0.51820844\n",
      "0.32400513\n",
      "0.768514\n",
      "0.6437023\n",
      "0.48822078\n"
     ]
    }
   ],
   "source": [
    "labels = [\"original\", \"summary (TextRank)\", \"summary (Centroid-BOW)\", \"summary (control)\"]\n",
    "gdict = {\"politics\": [], \"books\": [], \"movies\": [], \"sports\": []}\n",
    "gdictadd([mpol, mmov, mbooks, msports])\n",
    "gdictadd([mpol_s, mbks_s, mmov_s, mspts_s])\n",
    "gdictadd([mpol_b, mbks_b, mmov_b, mspts_b])\n",
    "gdictadd([mpol_c, mbks_c, mmov_c, mspts_c])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "16477"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "vocab = set(mpol.wv.vocab)\n",
    "for m in [mpol, mmov, mbooks, msports]:\n",
    "    vocab &= set(m.wv.vocab)\n",
    "len(vocab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 151,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "5573"
      ]
     },
     "execution_count": 151,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "vocab_all = set(vocab)\n",
    "for m in [mpol_s, mbks_s, mmov_s, mspts_s] + \\\n",
    "[mpol_b, mbks_b, mmov_b, mspts_b] + \\\n",
    "[mpol_c, mbks_c, mmov_c, mspts_c]:\n",
    "    vocab_all &= set(m.wv.vocab)\n",
    "len(vocab_all)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 152,
   "metadata": {},
   "outputs": [],
   "source": [
    "vocab2 = vocab_all"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [],
   "source": [
    "proflist = list(filter(lambda x: x in mbooks.wv.vocab, [p[0] for p in profs]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('debiaswe-master/data/equalize_pairs.json') as f:\n",
    "    pairs = json.loads(f.read())\n",
    "pairs = list(map(lambda x: [x[0].lower(), x[1].lower()], pairs))\n",
    "with open('debiaswe-master/data/professions.json') as f:\n",
    "    profs = json.loads(f.read())\n",
    "# proflist = list(filter(lambda x: x in mpol.wv.vocab, [p[0] for p in profs if abs(p[1]) < 0.6]))\n",
    "with open('debiaswe-master/data/gender_specific_seed.json') as f:\n",
    "    gender_specific = json.loads(f.read())\n",
    "with open('objects.csv') as csv_file:\n",
    "    objects = csv_file.read().split(\"\\n\")\n",
    "objects[0] = 'angle'\n",
    "with open('cities.csv') as csv_file:\n",
    "    cities = csv_file.read().split(\",\")\n",
    "cities = [x.lower() for x in cities]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [],
   "source": [
    "def getg(m):\n",
    "    matrix = []\n",
    "    for a, b in pairs:\n",
    "        if a not in m.wv.vocab or b not in m.wv.vocab:\n",
    "            continue\n",
    "        center = (m.wv.get_vector(a) + m.wv.get_vector(b))/2\n",
    "        matrix.append(m.wv.get_vector(a) - center)\n",
    "        matrix.append(m.wv.get_vector(b) - center)\n",
    "    matrix = np.array(matrix)\n",
    "    p = PCA(n_components = 10)\n",
    "    p.fit(matrix)\n",
    "    # print(p.explained_variance_ratio_[0])\n",
    "#         plt.bar(range(10), p.explained_variance_ratio_)\n",
    "#         plt.title(\"Explained Variance vs Component Number (\" + t + \")\")\n",
    "    return p.components_[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 135,
   "metadata": {},
   "outputs": [],
   "source": [
    "def gproj2(g1, g2, m, w1):\n",
    "    wvec = m.wv.get_vector(w1)\n",
    "    # g = project wvec onto g1, g2 subspace\n",
    "    A = np.hstack((g1.reshape(100,1),g2.reshape(100,1)))\n",
    "    P = A @ np.linalg.inv(A.T @ A) @ A.T\n",
    "    g = P @ wvec\n",
    "    p = np.dot(g, wvec)/(np.linalg.norm(g)*np.linalg.norm(wvec))\n",
    "    return p"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 136,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.3386617\n",
      "though          0.00 \t\t chairman        0.87\n",
      "proper          0.00 \t\t democrat        0.85\n",
      "translate       0.00 \t\t speaker         0.81\n",
      "combined        0.01 \t\t chief           0.79\n",
      "modernity       0.01 \t\t james           0.79\n",
      "fears           0.01 \t\t lawyer          0.78\n",
      "moral           0.01 \t\t jr              0.78\n",
      "excess          0.01 \t\t member          0.78\n",
      "meaningful      0.01 \t\t leader          0.78\n",
      "collection      0.01 \t\t secretary       0.77\n",
      "sport           0.01 \t\t representative  0.77\n",
      "dragging        0.01 \t\t john            0.76\n",
      "practices       0.01 \t\t david           0.76\n",
      "shifting        0.01 \t\t congressman     0.75\n",
      "expenses        0.01 \t\t deputy          0.75\n"
     ]
    }
   ],
   "source": [
    "gpol_s1, gpol_s2 = getg(mpol_s)\n",
    "res = sorted([(x, gproj2(gpol_s1, gpol_s2, mpol_s, x)) for x in filter(lambda x: x in vocab2, vocab2)], key=lambda x: x[1])\n",
    "for i in range(15):\n",
    "    print(f\"{res[i][0].ljust(15)} {res[i][1]:.2f} \\t\\t {res[-i-1][0].ljust(15)} {res[-i-1][1]:.2f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "def gdictadd(models):\n",
    "    for i in range(len(models)):\n",
    "        gdict[list(gdict)[i]].append(getg(models[i]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "def mdictadd(models):\n",
    "    for i in range(len(models)):\n",
    "        mdict[list(mdict)[i]].append(models[i])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "def gproj(g, m, w1):\n",
    "    wvec = m.wv.get_vector(w1)\n",
    "    p = np.dot(g, wvec)/(np.linalg.norm(g)*np.linalg.norm(wvec))\n",
    "    return p"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [],
   "source": [
    "def bias(g, m, vocab, wordlist):\n",
    "    b = 0\n",
    "    l = list(filter(lambda x: x in vocab, wordlist))\n",
    "    for x in l:\n",
    "        b += abs(gproj(g, m, x))\n",
    "    return b/len(l)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<gensim.models.word2vec.Word2Vec at 0x1188db518>,\n",
       " <gensim.models.word2vec.Word2Vec at 0x16d225c50>]"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [],
   "source": [
    "proflist = list(filter(lambda x: x in mbooks.wv.vocab, [p[0] for p in profs if abs(p[1]) < 0.6]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.34738272\n",
      "0.2716877\n",
      "    sports 0.298 0.164 0.266 0.231 0.334\n",
      "   sports2 0.289 0.149 0.269 0.204 0.324\n"
     ]
    }
   ],
   "source": [
    "gdict = {\"sports\": [], \"sports2\": []}\n",
    "mdict = {\"sports\": [], \"sports2\": []}\n",
    "models = [msports, msports1]\n",
    "gdictadd(models)\n",
    "mdictadd(models)\n",
    "for i in gdict:\n",
    "    print(\"%10s\" % i, \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, proflist), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, adjectives), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, cities), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, objects), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, gender_specific))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.34738272\n",
      "0.28547773\n",
      "0.23369507\n",
      "    sports 0.306 0.166 0.129 0.352\n",
      "   sports1 0.299 0.156 0.134 0.343\n",
      "   sports2 0.242 0.132 0.117 0.297\n"
     ]
    }
   ],
   "source": [
    "gdict = {\"sports\": [], \"sports1\": [], \"sports2\": []}\n",
    "mdict = {\"sports\": [], \"sports1\": [], \"sports2\": []}\n",
    "models = [msports, msports1, msports2]\n",
    "gdictadd(models)\n",
    "mdictadd(models)\n",
    "for i in gdict:\n",
    "    print(\"%10s\" % i, \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, proflist), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, adjectives), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, grammar), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, gender_specific))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.5087397\n",
      "0.40879804\n",
      "0.3250344\n",
      "       mov 0.085 0.087 0.039 0.172\n",
      "      mov1 0.094 0.083 0.040 0.184\n",
      "      mov2 0.096 0.082 0.046 0.188\n"
     ]
    }
   ],
   "source": [
    "gdict = {\"mov\": [], \"mov1\": [], \"mov2\": []}\n",
    "mdict = {\"mov\": [], \"mov1\": [], \"mov2\": []}\n",
    "models = [mmov, mmov1, mmov2]\n",
    "gdictadd(models)\n",
    "mdictadd(models)\n",
    "for i in gdict:\n",
    "    print(\"%10s\" % i, \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, proflist), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, adjectives), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, grammar), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, gender_specific))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.34950238\n",
      "0.3026881\n",
      "0.26836872\n",
      "       pol 0.091 0.070 0.053 0.125\n",
      "      pol1 0.128 0.077 0.057 0.139\n",
      "      pol2 0.111 0.079 0.055 0.144\n"
     ]
    }
   ],
   "source": [
    "gdict = {\"pol\": [], \"pol1\": [], \"pol2\": []}\n",
    "mdict = {\"pol\": [], \"pol1\": [], \"pol2\": []}\n",
    "models = [mpol, mpol1, mpol2]\n",
    "gdictadd(models)\n",
    "mdictadd(models)\n",
    "for i in gdict:\n",
    "    print(\"%10s\" % i, \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, proflist), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, adjectives), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, grammar), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, gender_specific))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.5940078\n",
      "0.48473203\n",
      "0.37103993\n",
      "     books 0.100 0.067 0.038 0.159\n",
      "    books1 0.123 0.081 0.038 0.178\n",
      "    books2 0.138 0.087 0.038 0.189\n"
     ]
    }
   ],
   "source": [
    "gdict = {\"books\": [], \"books1\": [], \"books2\": []}\n",
    "mdict = {\"books\": [], \"books1\": [], \"books2\": []}\n",
    "models = [mbooks, mbooks1, mbooks2]\n",
    "gdictadd(models)\n",
    "mdictadd(models)\n",
    "for i in gdict:\n",
    "    print(\"%10s\" % i, \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, proflist), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, adjectives), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, grammar), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, gender_specific))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#todo"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 158,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "original\n",
      "  politics 0.099 0.066 0.072 0.058 0.120\n",
      "     books 0.086 0.092 0.089 0.080 0.173\n",
      "    movies 0.096 0.069 0.077 0.065 0.155\n",
      "    sports 0.327 0.175 0.258 0.228 0.372\n"
     ]
    }
   ],
   "source": [
    "print(labels[0])\n",
    "for i in gdict:\n",
    "    print(\"%10s\" % i, \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, proflist), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, adjectives), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, cities), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, objects), \\\n",
    "          \"%.3f\" % bias(gdict[i][0], mdict[i][0], vocab, gender_specific))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 159,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "summary (TextRank)\n",
      "  politics 0.392 0.111 0.294 0.146 0.269\n",
      "     books 0.052 0.051 0.044 0.052 0.161\n",
      "    movies 0.083 0.089 0.064 0.084 0.192\n",
      "    sports 0.303 0.154 0.182 0.180 0.283\n"
     ]
    }
   ],
   "source": [
    "print(labels[1])\n",
    "for i in gdict:\n",
    "    print(\"%10s\" % i, \\\n",
    "          \"%.3f\" % bias(gdict[i][1], mdict[i][1], vocab2, proflist), \\\n",
    "          \"%.3f\" % bias(gdict[i][1], mdict[i][1], vocab2, adjectives), \\\n",
    "          \"%.3f\" % bias(gdict[i][1], mdict[i][1], vocab2, cities), \\\n",
    "          \"%.3f\" % bias(gdict[i][1], mdict[i][1], vocab2, objects), \\\n",
    "          \"%.3f\" % bias(gdict[i][1], mdict[i][1], vocab2, gender_specific))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 160,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "summary (Centroid-BOW)\n",
      "  politics 0.512 0.152 0.446 0.229 0.405\n",
      "     books 0.048 0.054 0.046 0.048 0.163\n",
      "    movies 0.069 0.077 0.067 0.088 0.194\n",
      "    sports 0.337 0.181 0.209 0.198 0.298\n"
     ]
    }
   ],
   "source": [
    "print(labels[2])\n",
    "for i in gdict:\n",
    "    print(\"%10s\" % i, \\\n",
    "          \"%.3f\" % bias(gdict[i][2], mdict[i][2], vocab2, proflist), \\\n",
    "          \"%.3f\" % bias(gdict[i][2], mdict[i][2], vocab2, adjectives), \\\n",
    "          \"%.3f\" % bias(gdict[i][2], mdict[i][2], vocab2, cities), \\\n",
    "          \"%.3f\" % bias(gdict[i][2], mdict[i][2], vocab2, objects), \\\n",
    "          \"%.3f\" % bias(gdict[i][2], mdict[i][2], vocab2, gender_specific))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 161,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "summary (control)\n",
      "  politics 0.183 0.091 0.143 0.090 0.123\n",
      "     books 0.046 0.045 0.037 0.049 0.137\n",
      "    movies 0.066 0.069 0.051 0.068 0.172\n",
      "    sports 0.246 0.115 0.145 0.135 0.197\n"
     ]
    }
   ],
   "source": [
    "print(labels[3])\n",
    "for i in gdict:\n",
    "    print(\"%10s\" % i, \\\n",
    "          \"%.3f\" % bias(gdict[i][3], mdict[i][3], vocab2, proflist), \\\n",
    "          \"%.3f\" % bias(gdict[i][3], mdict[i][3], vocab2, adjectives), \\\n",
    "          \"%.3f\" % bias(gdict[i][3], mdict[i][3], vocab2, cities), \\\n",
    "          \"%.3f\" % bias(gdict[i][3], mdict[i][3], vocab2, objects), \\\n",
    "          \"%.3f\" % bias(gdict[i][3], mdict[i][3], vocab2, gender_specific))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 162,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.35275307\n",
      "fielding   -0.48 \t\t ms         0.66\n",
      "consistently -0.44 \t\t mrs        0.65\n",
      "solid      -0.43 \t\t husband    0.64\n",
      "shaky      -0.40 \t\t sang       0.62\n",
      "switch     -0.40 \t\t house      0.58\n",
      "hitting    -0.39 \t\t local      0.58\n",
      "throwing   -0.38 \t\t store      0.58\n",
      "hernández  -0.38 \t\t sister     0.57\n",
      "crucial    -0.38 \t\t ann        0.56\n",
      "starting   -0.37 \t\t songs      0.56\n"
     ]
    }
   ],
   "source": [
    "gsports = getg(msports)\n",
    "res = sorted([(x, gproj(gsports, msports, x)) for x in filter(lambda x: x in vocab, vocab)], key=lambda x: x[1])\n",
    "for i in range(10):\n",
    "    print(f\"{res[i][0].ljust(10)} {res[i][1]:.2f} \\t\\t {res[-i-1][0].ljust(10)} {res[-i-1][1]:.2f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.34343678\n",
      "her        -0.50 \t\t mr         0.43\n",
      "women      -0.47 \t\t ouster     0.39\n",
      "ms         -0.46 \t\t abbas      0.39\n",
      "she        -0.46 \t\t mikhail    0.38\n",
      "feeding    -0.37 \t\t mahmoud    0.37\n",
      "disabled   -0.35 \t\t arafat     0.37\n",
      "herself    -0.35 \t\t vladimir   0.37\n",
      "mental     -0.34 \t\t himself    0.36\n",
      "female     -0.34 \t\t kremlin    0.36\n",
      "classroom  -0.32 \t\t hamid      0.35\n"
     ]
    }
   ],
   "source": [
    "gpol = getg(mpol)\n",
    "res = sorted([(x, gproj(gpol, mpol, x)) for x in filter(lambda x: x in vocab, vocab)], key=lambda x: x[1])\n",
    "for i in range(10):\n",
    "    print(f\"{res[i][0].ljust(10)} {res[i][1]:.2f} \\t\\t {res[-i-1][0].ljust(10)} {res[-i-1][1]:.2f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.62817353\n",
      "her        -0.57 \t\t disney     0.41\n",
      "herself    -0.47 \t\t pictures   0.38\n",
      "she        -0.44 \t\t production 0.35\n",
      "husband    -0.41 \t\t studio     0.34\n",
      "mother     -0.39 \t\t mr         0.34\n",
      "ms         -0.38 \t\t producers  0.34\n",
      "girl       -0.38 \t\t kong       0.34\n",
      "woman      -0.36 \t\t paramount  0.33\n",
      "daughter   -0.35 \t\t entertainment 0.33\n",
      "single     -0.34 \t\t universal  0.32\n"
     ]
    }
   ],
   "source": [
    "gmov_s = getg(mmov_s)\n",
    "res = sorted([(x, gproj(gmov_s, mmov_s, x)) for x in filter(lambda x: x in vocab2, vocab2)], key=lambda x: x[1])\n",
    "for i in range(10):\n",
    "    print(f\"{res[i][0].ljust(10)} {res[i][1]:.2f} \\t\\t {res[-i-1][0].ljust(10)} {res[-i-1][1]:.2f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 119,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.78374213\n",
      "mr         -0.38 \t\t her        0.63\n",
      "he         -0.32 \t\t she        0.47\n",
      "his        -0.30 \t\t herself    0.46\n",
      "administration -0.28 \t\t husband    0.42\n",
      "iraq       -0.27 \t\t baby       0.37\n",
      "president  -0.26 \t\t girl       0.35\n",
      "military   -0.26 \t\t children   0.31\n",
      "bush       -0.26 \t\t mother     0.30\n",
      "nations    -0.26 \t\t sister     0.30\n",
      "history    -0.24 \t\t love       0.29\n"
     ]
    }
   ],
   "source": [
    "gbks_s = getg(mbks_s)\n",
    "res = sorted([(x, gproj(gbks_s, mbks_s, x)) for x in filter(lambda x: x in vocab2, vocab2)], key=lambda x: x[1])\n",
    "for i in range(10):\n",
    "    print(f\"{res[i][0].ljust(10)} {res[i][1]:.2f} \\t\\t {res[-i-1][0].ljust(10)} {res[-i-1][1]:.2f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.3386617\n",
      "chairman        -0.87 \t\t these           0.47\n",
      "democrat        -0.82 \t\t large           0.46\n",
      "speaker         -0.81 \t\t low             0.46\n",
      "chief           -0.77 \t\t access          0.45\n",
      "leader          -0.77 \t\t illegal         0.44\n",
      "member          -0.77 \t\t prices          0.43\n",
      "jr              -0.76 \t\t pay             0.42\n",
      "secretary       -0.76 \t\t without         0.42\n",
      "james           -0.76 \t\t oil             0.41\n",
      "lawyer          -0.76 \t\t making          0.41\n",
      "deputy          -0.75 \t\t flow            0.41\n",
      "representative  -0.74 \t\t buy             0.41\n",
      "richard         -0.74 \t\t real            0.40\n",
      "aide            -0.74 \t\t raise           0.40\n",
      "david           -0.73 \t\t stock           0.39\n"
     ]
    }
   ],
   "source": [
    "gpol_s = getg(mpol_s)\n",
    "res = sorted([(x, gproj(gpol_s, mpol_s, x)) for x in filter(lambda x: x in vocab2, vocab2)], key=lambda x: x[1])\n",
    "for i in range(15):\n",
    "    print(f\"{res[i][0].ljust(15)} {res[i][1]:.2f} \\t\\t {res[-i-1][0].ljust(15)} {res[-i-1][1]:.2f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.32400513\n",
      "mail       -0.37 \t\t head       0.52\n",
      "or         -0.35 \t\t chairman   0.51\n",
      "high       -0.35 \t\t union      0.51\n",
      "got        -0.33 \t\t president  0.50\n",
      "few        -0.32 \t\t leadership 0.48\n",
      "com        -0.32 \t\t secretary  0.48\n",
      "young      -0.31 \t\t chief      0.47\n",
      "children   -0.31 \t\t powers     0.47\n",
      "isn        -0.29 \t\t coalition  0.47\n",
      "these      -0.28 \t\t leader     0.47\n"
     ]
    }
   ],
   "source": [
    "gpol_c = getg(mpol_c)\n",
    "res = sorted([(x, gproj(gpol_c, mpol_c, x)) for x in filter(lambda x: x in vocab2, vocab2)], key=lambda x: x[1])\n",
    "for i in range(10):\n",
    "    print(f\"{res[i][0].ljust(10)} {res[i][1]:.2f} \\t\\t {res[-i-1][0].ljust(10)} {res[-i-1][1]:.2f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.768514\n",
      "her        -0.61 \t\t mr         0.38\n",
      "herself    -0.46 \t\t his        0.31\n",
      "she        -0.46 \t\t war        0.29\n",
      "husband    -0.41 \t\t he         0.28\n",
      "children   -0.30 \t\t bush       0.27\n",
      "love       -0.28 \t\t civil      0.24\n",
      "ms         -0.27 \t\t administration 0.23\n",
      "mother     -0.27 \t\t ii         0.23\n",
      "girl       -0.27 \t\t iraq       0.22\n",
      "woman      -0.27 \t\t policy     0.20\n"
     ]
    }
   ],
   "source": [
    "gbks_c = getg(mbks_c)\n",
    "res = sorted([(x, gproj(gbks_c, mbks_c, x)) for x in filter(lambda x: x in vocab2, vocab2)], key=lambda x: x[1])\n",
    "for i in range(10):\n",
    "    print(f\"{res[i][0].ljust(10)} {res[i][1]:.2f} \\t\\t {res[-i-1][0].ljust(10)} {res[-i-1][1]:.2f}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 165,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, 'Explained Variance vs Component Number (Random)')"
      ]
     },
     "execution_count": 165,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "pca_r = PCA(n_components = 10)\n",
    "pca_r.fit(np.random.rand(70,100))\n",
    "plt.bar(range(10), pca_r.explained_variance_ratio_)\n",
    "plt.title(\"Explained Variance vs Component Number (Random)\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "gpol = getg(mpol, \"Gender Pairs - Politics\", True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Regression for bias testing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 189,
   "metadata": {},
   "outputs": [],
   "source": [
    "def regress(g1, m1, g2, m2, wordlist):\n",
    "    x = np.array([gproj(g1, m1, w) for w in wordlist if w in m1.wv.vocab and w in m2.wv.vocab])\n",
    "    y = np.array([gproj(g2, m2, w) for w in wordlist if w in m1.wv.vocab and w in m2.wv.vocab])\n",
    "    slope, intercept, r_value, p_value, std_err = stats.linregress(x,y)\n",
    "    return slope, p_value"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 190,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.31522945\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.7564479357792341, 1.1464826034410933e-07)"
      ]
     },
     "execution_count": 190,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "regress(gpol, mpol, getg(mpol_s), mpol_s, proflist)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}