SVM_model.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Import libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd \n",
    "import string\n",
    "import re \n",
    "import nltk\n",
    "import emoji\n",
    "import numpy\n",
    "import time\n",
    "from sklearn.metrics import f1_score\n",
    "from sklearn.metrics import make_scorer\n",
    "from sklearn.feature_extraction.text import CountVectorizer\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from sklearn import model_selection, svm\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.metrics import classification_report, confusion_matrix,f1_score, accuracy_score\n",
    "from sklearn.metrics import confusion_matrix\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from nltk.tokenize import word_tokenize\n",
    "from nltk.stem.porter import PorterStemmer\n",
    "from sklearn.pipeline import Pipeline\n",
    "from sklearn.pipeline import make_pipeline\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "import np\n",
    "from collections import Counter\n",
    "from textblob import TextBlob \n",
    "from langdetect import detect"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Function to generate confusion matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def make_confusion_matrix(cf, group_names=None, categories='auto', count=True, percent=True, cbar=True, xyticks=True,\n",
    "                          xyplotlabels=True,\n",
    "                          sum_stats=True,\n",
    "                          figsize=None,\n",
    "                          cmap='Blues',\n",
    "                          title=None):\n",
    "\n",
    "    # CODE TO GENERATE TEXT INSIDE EACH SQUARE\n",
    "    blanks = ['' for i in range(cf.size)]\n",
    "\n",
    "    if group_names and len(group_names)==cf.size:\n",
    "        group_labels = [\"{}\\n\".format(value) for value in group_names]\n",
    "    else:\n",
    "        group_labels = blanks\n",
    "\n",
    "    if count:\n",
    "        group_counts = [\"{0:0.0f}\\n\".format(value) for value in cf.flatten()]\n",
    "    else:\n",
    "        group_counts = blanks\n",
    "\n",
    "    if percent:\n",
    "        group_percentages = [\"{0:.2%}\".format(value) for value in cf.flatten()/np.sum(cf)]\n",
    "    else:\n",
    "        group_percentages = blanks\n",
    "\n",
    "    box_labels = [f\"{v1}{v2}{v3}\".strip() for v1, v2, v3 in zip(group_labels,group_counts,group_percentages)]\n",
    "    box_labels = np.asarray(box_labels).reshape(cf.shape[0],cf.shape[1])\n",
    "\n",
    "    # SET FIGURE PARAMETERS ACCORDING TO OTHER ARGUMENTS\n",
    "    if figsize==None:\n",
    "        #Get default figure size if not set\n",
    "        figsize = plt.rcParams.get('figure.figsize')\n",
    "\n",
    "    if xyticks==False:\n",
    "        #Do not show categories if xyticks is False\n",
    "        categories=False\n",
    "\n",
    "    # MAKE THE HEATMAP VISUALIZATION\n",
    "    plt.figure(figsize=figsize)\n",
    "    sns.heatmap(cf,annot=box_labels,fmt=\"\",cmap=cmap,cbar=cbar,xticklabels=categories,yticklabels=categories)\n",
    "    plt.ylabel('True label')\n",
    "    plt.xlabel('Predicted label')\n",
    "    plt.savefig('conf.pdf')  \n",
    "\n",
    "    if title:\n",
    "        plt.title(title)\n",
    "        "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Read, clean, prepare dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>tweetText</th>\n",
       "      <th>label</th>\n",
       "      <th>preprocessed</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>¿Se acuerdan de la película: “El día después d...</td>\n",
       "      <td>fake</td>\n",
       "      <td>acuerdan película día después mañana recuerda ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>@milenagimon: Miren a Sandy en NY!  Tremenda i...</td>\n",
       "      <td>fake</td>\n",
       "      <td>miren tremenda imagen parece día independencia...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Buena la foto del Huracán Sandy, me recuerda a...</td>\n",
       "      <td>fake</td>\n",
       "      <td>buena foto recuerda película día independencia...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Scary shit #hurricane #NY http://t.co/e4JLBUfH</td>\n",
       "      <td>fake</td>\n",
       "      <td>scary shit</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>My fave place in the world #nyc #hurricane #sa...</td>\n",
       "      <td>fake</td>\n",
       "      <td>fave place world</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                           tweetText label  \\\n",
       "0  ¿Se acuerdan de la película: “El día después d...  fake   \n",
       "1  @milenagimon: Miren a Sandy en NY!  Tremenda i...  fake   \n",
       "2  Buena la foto del Huracán Sandy, me recuerda a...  fake   \n",
       "3     Scary shit #hurricane #NY http://t.co/e4JLBUfH  fake   \n",
       "4  My fave place in the world #nyc #hurricane #sa...  fake   \n",
       "\n",
       "                                        preprocessed  \n",
       "0  acuerdan película día después mañana recuerda ...  \n",
       "1  miren tremenda imagen parece día independencia...  \n",
       "2  buena foto recuerda película día independencia...  \n",
       "3                                         scary shit  \n",
       "4                                   fave place world  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Read dataset\n",
    "test_tweets = pd.read_csv('mediaeval-2015-testset.txt', sep='\\\\t', engine='python', encoding='utf-8')\n",
    "train_tweets = pd.read_csv('mediaeval-2015-trainingset.txt', sep='\\\\t', engine='python', encoding='utf-8')\n",
    "\n",
    "def removeColumns(tweets):\n",
    "    tweets.drop([\"timestamp\"],axis=1,inplace=True)\n",
    "    tweets.drop([\"username\"],axis=1,inplace=True)\n",
    "    tweets.drop([\"tweetId\"],axis=1,inplace=True)\n",
    "    tweets.drop([\"userId\"],axis=1,inplace=True)\n",
    "    tweets.drop([\"imageId(s)\"],axis=1,inplace=True)\n",
    "\n",
    "# Remove Columns\n",
    "removeColumns(train_tweets)\n",
    "removeColumns(test_tweets)\n",
    "\n",
    "def converHumourLabel(tweets):\n",
    "    tweets.loc[tweets['label'] == 'humor', 'label'] = 'fake'\n",
    "    \n",
    "# Change 'humor' to 'fake'\n",
    "converHumourLabel(train_tweets)\n",
    "converHumourLabel(test_tweets)\n",
    "\n",
    "\n",
    "def englishTweets(tweets):\n",
    "    # Generate lang\n",
    "    ls = [] \n",
    "    for value in tweets[\"preprocessed\"]: \n",
    "        if not (value == \"en\"):\n",
    "            ls.append(detect(str(tweets)))\n",
    "        else:\n",
    "            ls.append(\"ND\")\n",
    "    tweets[\"lang\"] = ls \n",
    "    return tweets\n",
    "\n",
    "def clean(text):\n",
    "    # Remove URLs\n",
    "    tweet =  re.sub(r'http\\S+', '', text)\n",
    "    # Remove mentions\n",
    "    tweet = re.sub(r'@\\w+', '', tweet)\n",
    "    # Remove all the special characters\n",
    "    tweet = re.sub(r'\\W', ' ', tweet)\n",
    "    # Convert to lower case\n",
    "    tweet = tweet.lower()\n",
    "    return tweet\n",
    "\n",
    "\n",
    "# Remove stop words\n",
    "def remove_stopwords(text):\n",
    "    event_words = ['hurricane', 'sochi',\n",
    "                   'soldier', 'hurricanesandy', \n",
    "                   'liberty', 'bringbackourgirls', 'jersey', 'manhattan','flood',\n",
    "                   'nj', 'new', 'nyc', 'ny', 'york', 'statue',\n",
    "                   'statueofliberty', 'shark', 'newyork', \n",
    "                   'tomb',  'mh370', 'sandy', 'huracán', 'boston',  \n",
    "                   'columbianchemicals',\n",
    "                    'flooding', 'cuba']\n",
    "    \n",
    "    twitter = ['via', 'photo', 'rt']\n",
    "    en_stop_words = nltk.corpus.stopwords.words('english')\n",
    "    sp_stop_words = nltk.corpus.stopwords.words('spanish')\n",
    "    stopword = en_stop_words + sp_stop_words + twitter + event_words\n",
    "    text= ' '.join([word for word in text.split() if word not in stopword])\n",
    "    return text\n",
    "\n",
    "def stemming(text):\n",
    "    tokens = word_tokenize(text)\n",
    "    stemmer = PorterStemmer()\n",
    "    stems = [stemmer.stem(w) for w in tokens]\n",
    "    return ' '.join(stems)\n",
    "\n",
    "# Clean text\n",
    "train_tweets['preprocessed'] = train_tweets['tweetText'].apply(clean)\n",
    "test_tweets['preprocessed'] = test_tweets['tweetText'].apply(clean)\n",
    "\n",
    "# train_tweets = englishTweets()\n",
    "\n",
    "    \n",
    "# Remove stopwords\n",
    "train_tweets['preprocessed'] = train_tweets['preprocessed'].apply(remove_stopwords)\n",
    "test_tweets['preprocessed'] = test_tweets['preprocessed'].apply(remove_stopwords)\n",
    "\n",
    "# Perform stemming - ineffective\n",
    "# train_tweets['preprocessed'] = train_tweets['preprocessed'].apply(stemming)\n",
    "# test_tweets['preprocessed'] = test_tweets['preprocessed'].apply(stemming)\n",
    "\n",
    "\n",
    "\n",
    "test_tweets.to_csv('test_tweets.csv')\n",
    "train_tweets.to_csv('train_tweets.csv')\n",
    "\n",
    "train_tweets.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Hyperparameter tuning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Performing grid search...\n",
      "pipeline: ['tfidf', 'clf']\n",
      "parameters:\n",
      "{'tfidf__max_features': (3600, 10000)}\n",
      "Fitting 10 folds for each of 2 candidates, totalling 20 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 12 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done  18 out of  20 | elapsed:   14.4s remaining:    1.6s\n",
      "[Parallel(n_jobs=-1)]: Done  20 out of  20 | elapsed:   14.5s finished\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "done in 19.590s\n",
      "\n",
      "Best score: 0.846\n",
      "Best parameters set:\n",
      "\ttfidf__max_features: 3600\n"
     ]
    }
   ],
   "source": [
    "# Define x and y train and test set\n",
    "x_train = train_tweets['preprocessed']\n",
    "x_test = test_tweets['preprocessed']\n",
    "y_train = train_tweets['label']\n",
    "y_test = test_tweets['label']\n",
    "\n",
    "# classifier\n",
    "pipeline = Pipeline([\n",
    "    ('tfidf', TfidfVectorizer(max_df=0.5, min_df=2, ngram_range=(1,1))),\n",
    "    ('clf', svm.SVC(kernel='linear',C=1.9, gamma='auto')),\n",
    "])\n",
    "\n",
    "parameters = {\n",
    "    'tfidf__max_features': (3600,10000)\n",
    "}\n",
    "\n",
    "f1_scorer = make_scorer(f1_score, average='micro', pos_label=\"fake\")\n",
    "\n",
    "# find the best parameters for both the feature extraction and the\n",
    "# classifier\n",
    "grid_search = GridSearchCV(pipeline, parameters, n_jobs=-1, cv=10 , verbose=1,scoring=f1_scorer)\n",
    "\n",
    "print(\"Performing grid search...\")\n",
    "print(\"pipeline:\", [name for name, _ in pipeline.steps])\n",
    "print(\"parameters:\")\n",
    "print(parameters)\n",
    "t0 = time.time()\n",
    "grid_search.fit(x_train, y_train)\n",
    "print(\"done in %0.3fs\" % (time.time() - t0))\n",
    "print()\n",
    "print(\"Best score: %0.3f\" % grid_search.best_score_)\n",
    "print(\"Best parameters set:\")\n",
    "best_parameters = grid_search.best_estimator_.get_params()\n",
    "for param_name in sorted(parameters.keys()):\n",
    "    print(\"\\t%s: %r\" % (param_name, best_parameters[param_name]))\n",
    "    \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "micro F1 score: 0.8508331129330866\n"
     ]
    }
   ],
   "source": [
    "predictions = grid_search.predict(x_test)\n",
    "print(\"micro F1 score: \" + str(f1_score(y_test, predictions, average='micro')))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train and evaluate model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "micro F1-score: 0.8804549061094947\n",
      "macro F1-score: 0.8589963182538833\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Define x and y train and test set\n",
    "x_train = train_tweets['preprocessed']\n",
    "x_test = test_tweets['preprocessed']\n",
    "y_train = train_tweets['label']\n",
    "y_test = test_tweets['label']\n",
    "\n",
    "# TF-IDF vectorizer TfidfVectorizer()\n",
    "# Tfidf_vect = TfidfVectorizer(max_features=3600, min_df=2, max_df=0.5, ngram_range=(1,1))\n",
    "# Tfidf_vect = TfidfVectorizer()\n",
    "\n",
    "# unigrams performs very well\n",
    "# Tfidf_vect = CountVectorizer(max_features=3600, min_df=2, ngram_range=(1,1)) \n",
    "\n",
    "Tfidf_vect = TfidfVectorizer(max_features=3600, min_df=2, max_df=0.5, ngram_range=(1,1))\n",
    "\n",
    "\n",
    "Tfidf_vect.fit(train_tweets['preprocessed'])\n",
    "train_x_tfidf = Tfidf_vect.transform(x_train)\n",
    "test_x_tfidf = Tfidf_vect.transform(x_test)\n",
    "\n",
    "labels = ['True Neg','False Pos','False Neg','True Pos']\n",
    "categories = ['Fake', 'Real']\n",
    "\n",
    "# Classifier - Support Vector Machine\n",
    "SVM = svm.SVC(C=1.9, kernel='linear')\n",
    "SVM.fit(train_x_tfidf, y_train)\n",
    "svm_predictions = SVM.predict(test_x_tfidf)\n",
    "svm_cf_matrix = confusion_matrix(y_test, svm_predictions)\n",
    "\n",
    "make_confusion_matrix(svm_cf_matrix, group_names=labels, categories=categories, cmap='Blues')\n",
    "print(\"micro F1-score: \" + str(f1_score(y_test, svm_predictions, average='micro')))\n",
    "print(\"macro F1-score: \" + str(f1_score(y_test, svm_predictions, average='macro')))\n",
    "\n"
   ]
  }
 ],
 "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": "3.8.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}