get_similar_papers.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "74c09945-0977-4124-834c-6fbf5c31c86d",
   "metadata": {},
   "source": [
    "### This file will extract a list of paper DOIs that belong to similar papers to a given seed paper.\n",
    "- Similarity is measured using sentence embeddings. Similarity threshold can be decided manually\n",
    "- Make sure to execute every cell sequentially. get_similar_papers() is the function that returns a list of relevant papers' DOIs. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "a73d16fc-1f75-418e-ae54-c87cee4d6bee",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/bowenyi/.local/lib/python3.11/site-packages/sentence_transformers/cross_encoder/CrossEncoder.py:13: TqdmExperimentalWarning: Using `tqdm.autonotebook.tqdm` in notebook mode. Use `tqdm.tqdm` instead to force console mode (e.g. in jupyter console)\n",
      "  from tqdm.autonotebook import tqdm, trange\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "device = 'cuda' if torch.cuda.is_available() else 'cpu'\n",
    "\n",
    "import json\n",
    "from sentence_transformers import SentenceTransformer, util\n",
    "import numpy as np\n",
    "import random\n",
    "random.seed(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "99cc20d3-8586-44b2-bdc0-d6e1e5496c75",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/bowenyi/.local/lib/python3.11/site-packages/huggingface_hub/file_download.py:1132: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.\n",
      "  warnings.warn(\n"
     ]
    }
   ],
   "source": [
    "model = SentenceTransformer('all-mpnet-base-v2').to(device)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "19993d38-f81b-4bbd-afdd-83b866aba918",
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(\"seed_paper_10.json\") as file:  # Take seed paper 10 as example\n",
    "    refs_10 = json.load(file) \n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "id": "6408e564-5e9c-44de-bbef-f131c99416cb",
   "metadata": {},
   "outputs": [],
   "source": [
    "def is_similar(target, ref, threshold):  # Use sentence embeddings to measure the similarity and return a similarity score\n",
    "    target_embedding = model.encode(target, convert_to_tensor=True, device=device)\n",
    "    ref_embeddings = model.encode(ref, convert_to_tensor=True, device=device)\n",
    "    similarity_scores = model.similarity(target_embedding, ref_embeddings)\n",
    "    return similarity_scores[0] >= threshold, similarity_scores[0]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "49e060f8-8aee-4b6c-bb75-60427e74e9fd",
   "metadata": {},
   "outputs": [],
   "source": [
    "def concatenate_title_abs(title, abs):  # Concatenate title with abstract in account of type mismatch\n",
    "    if type(title) != str or type(abs) != str:\n",
    "        if type(title) == list:\n",
    "            str_title = \"\"\n",
    "            for text in title:\n",
    "                str_title += text + \" \"\n",
    "            title = str_title\n",
    "        \n",
    "        if type(abs) == list:\n",
    "            str_abs = \"\"\n",
    "            for text in abs:\n",
    "                str_abs += text + \" \"\n",
    "            abs = str_abs\n",
    "        \n",
    "            \n",
    "    return str(title) + \": \" + str(abs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "id": "2f0850f0-fdb3-41c6-9138-05b0a2198f22",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Return (1) a list of DOIs from similar papers to a seed paper and (2) total number of papers\n",
    "# Sample usage: similar_papers_8, paper_count_8 = get_similar_papers(seed_paper_refs=refs_8)\n",
    "\n",
    "def get_similar_papers(seed_paper_refs, similarity_threshold=0.64):  \n",
    "    similar_papers = []\n",
    "    paper_count = 1\n",
    "    \n",
    "    seed_title = seed_paper_refs[0]['metadata']['title']\n",
    "    seed_abs = seed_paper_refs[0]['metadata']['abstract']\n",
    "    seed_title_abs =  concatenate_title_abs(seed_title, seed_abs) \n",
    "    \n",
    "    for lev_1_ref in seed_paper_refs[0]['references']:\n",
    "        lev_1_title = lev_1_ref['metadata']['title']\n",
    "        lev_1_abs = lev_1_ref['metadata']['abstract']\n",
    "        lev_1_title_abs = concatenate_title_abs(lev_1_title, lev_1_abs)  \n",
    "        similar_lev_1, score_lev_1 = is_similar(seed_title_abs, lev_1_title_abs, similarity_threshold)\n",
    "        \n",
    "        paper_count += 1 + len(lev_1_ref['references'])\n",
    "        \n",
    "        \n",
    "        if similar_lev_1 and len(lev_1_ref['metadata']['doi']) != 0:\n",
    "            similar_papers.append(lev_1_ref['metadata']['doi'])\n",
    "            # similar_papers.append((lev_1_ref['metadata']['doi'], score_lev_1))\n",
    "\n",
    "            for lev_2_ref in lev_1_ref['references']:\n",
    "                lev_2_title = lev_2_ref['metadata']['title']\n",
    "                lev_2_abs = lev_2_ref['metadata']['abstract']\n",
    "                lev_2_title_abs = concatenate_title_abs(lev_2_title, lev_2_abs)\n",
    "                \n",
    "                target_lev_2 = seed_title_abs + \" \" + lev_1_title_abs\n",
    "                similar_lev_2, score_lev_2 = is_similar(target_lev_2, lev_2_title_abs, similarity_threshold)\n",
    "                \n",
    "                \n",
    "                if similar_lev_2 and len(lev_2_ref['metadata']['doi']) != 0:\n",
    "                    similar_papers.append(lev_2_ref['metadata']['doi'])\n",
    "                    # similar_papers.append((lev_2_ref['metadata']['doi'], score_lev_2))\n",
    "    \n",
    "\n",
    "    return similar_papers, paper_count\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "id": "267ea957-37d3-4e99-9778-4aac55650f37",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Sample usage:\n",
    "similar_papers_10, paper_count_10 = get_similar_papers(seed_paper_refs=refs_10)\n",
    "print(len(similar_papers_10))\n",
    "print(paper_count_10)\n",
    "print(similar_papers_10[0:10])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "487d1fc9-e506-4ec7-a3d5-26622ba7fa44",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.11.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}