Upload benchmark.ipynb

benchmark.ipynb

@@ -0,0 +1,364 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "e30bf074",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2023-06-14T17:16:17.863326Z",
+     "start_time": "2023-06-14T17:16:17.806871Z"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "import jellyfish\n",
+    "import nltk\n",
+    "import statistics\n",
+    "from collections import Counter\n",
+    "from tqdm import tqdm"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "711902a2",
+   "metadata": {},
+   "source": [
+    "# Load KWX"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "2bba73c8",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2023-06-14T17:12:46.170356Z",
+     "start_time": "2023-06-14T17:12:44.665181Z"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "kwx = json.load(open('./data.json','r'))\n",
+    "train_size = 16000"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "ed39b665",
+   "metadata": {},
+   "source": [
+    "# Metrics"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "b7df88f3",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2023-06-14T17:19:12.956834Z",
+     "start_time": "2023-06-14T17:19:12.949734Z"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "def compare_two_words(w1,w2):\n",
+    "    return jellyfish.jaro_distance(w1, w2)\n",
+    "\n",
+    "def is_related_in_list(word,ys):\n",
+    "    for y in ys:\n",
+    "        if compare_two_words(word.lower(),y.lower()) >= 0.9:\n",
+    "            return True\n",
+    "    return False \n",
+    "\n",
+    "def kws_precision(ys,ys_true):\n",
+    "    if len(ys) >= 1:\n",
+    "        count = 0 \n",
+    "        for y in ys:\n",
+    "            if is_related_in_list(y,ys_true):\n",
+    "                count += 1\n",
+    "        return count/len(ys)\n",
+    "    else:\n",
+    "        return 0\n",
+    "\n",
+    "def kws_recall(ys,ys_true):\n",
+    "    if not ys_true:\n",
+    "        return 0\n",
+    "    if len(ys) >= 1:\n",
+    "        count = 0 \n",
+    "        for y in ys:\n",
+    "            if is_related_in_list(y,ys_true):\n",
+    "                count += 1\n",
+    "        if count > len(ys_true):\n",
+    "            return 1\n",
+    "        return count/len(ys_true)\n",
+    "    else:\n",
+    "        return 0 \n",
+    "    \n",
+    "def evaluate_kws(ys,ys_true):\n",
+    "    res = {}\n",
+    "    res['precision'] = kws_precision(ys,ys_true)\n",
+    "    res['recall'] = kws_recall(ys,ys_true)\n",
+    "    if res['precision'] or res['recall']:\n",
+    "        res['f1'] = (2*res['precision']*res['recall'])/(res['precision']+res['recall'])\n",
+    "    else:\n",
+    "        res['f1'] = 0\n",
+    "    return res \n",
+    "\n",
+    "\n",
+    "def macro_pr(yt_pairs):\n",
+    "    f1 = []\n",
+    "    p = []\n",
+    "    r = []\n",
+    "    for pair in yt_pairs:\n",
+    "        evaluate = evaluate_kws(pair[0],pair[1])\n",
+    "        f1.append(evaluate['f1'])\n",
+    "        p.append(evaluate['precision'])\n",
+    "        r.append(evaluate['recall'])\n",
+    "    return {'f1':statistics.mean(f1),\n",
+    "           'precision':statistics.mean(p),\n",
+    "           'recall':statistics.mean(r)}\n",
+    "\n",
+    "def evaluate_extractor(docs_with_kws, extractor, top=None):\n",
+    "    yt_pairs = []\n",
+    "    for i in tqdm(list(docs_with_kws.keys())[train_size:]):\n",
+    "        pair = []\n",
+    "        pair.append(extractor(docs_with_kws[i]['abstract'],top))\n",
+    "        ys_true = docs_with_kws[i]['keywords']\n",
+    "        if 'keywords_extra' in docs_with_kws[i]:\n",
+    "            ys_true += docs_with_kws[i]['keywords_extra']\n",
+    "        pair.append(ys_true)\n",
+    "        yt_pairs.append(pair)\n",
+    "    print(len(yt_pairs))\n",
+    "    print(macro_pr(yt_pairs))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "949cd440",
+   "metadata": {},
+   "source": [
+    "# Keyword Extraction"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "30cd3512",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2023-06-14T17:12:51.826420Z",
+     "start_time": "2023-06-14T17:12:51.821688Z"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "kw_start = [\n",
+    "    'NN',\n",
+    "    'NNP',\n",
+    "    'JJ',\n",
+    "    'NNS',\n",
+    "    'VBG',\n",
+    "    'RB',\n",
+    "    'VBN'\n",
+    "]\n",
+    "\n",
+    "kw_end = [\n",
+    "    'NN',\n",
+    "    'NNP',\n",
+    "    'NNS',\n",
+    "    'VBG',\n",
+    "    'JJ'\n",
+    "]\n",
+    "\n",
+    "kw_split = [\n",
+    "    '.',\n",
+    "    ','\n",
+    "]\n",
+    "def keywords_selector(text):\n",
+    "    tokens = nltk.word_tokenize(text)\n",
+    "    tags = nltk.pos_tag(tokens)\n",
+    "    res = set()\n",
+    "    for i in range(len(tags)):\n",
+    "        if tags[i][1] in kw_start:\n",
+    "            end = i+4 if i+4 <= len(tags) - 1 else len(tags) - 1\n",
+    "            cut = tags[i:end]\n",
+    "            for j in range(len(cut)):\n",
+    "                if cut[j][0] in kw_split:\n",
+    "                    cut = cut[:j]\n",
+    "                    break \n",
+    "            for k in range(len(cut)):\n",
+    "                if cut[k][1] in kw_end:\n",
+    "                    res.add(' '.join([m[0] for m in cut][:k+1]))\n",
+    "    return res \n",
+    "    "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "00bf7e61",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2023-06-14T17:14:54.344161Z",
+     "start_time": "2023-06-14T17:13:30.555540Z"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "# scoreboard to record all combinations.\n",
+    "history_score = {}\n",
+    "# global dictionary to record all links between present and absent keywords\n",
+    "global_kwx_index = {}\n",
+    "for i in tqdm(list(kwx.keys())[:train_size]):\n",
+    "    ys = kwx[i]['keywords']\n",
+    "    candidates = keywords_selector(kwx[i]['abstract'])\n",
+    "    for c in candidates:\n",
+    "        if not c in history_score:\n",
+    "            history_score[c] = 0\n",
+    "        if is_related_in_list(c,ys):\n",
+    "            history_score[c] += 100\n",
+    "        else:\n",
+    "            history_score[c] -= 1\n",
+    "    for y in ys:\n",
+    "        if y in global_kwx_index:\n",
+    "            global_kwx_index[y].update(ys)\n",
+    "        else:\n",
+    "            global_kwx_index[y] = Counter(ys)\n",
+    "print(len(history_score))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "cf0bc909",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2023-06-14T17:20:03.965152Z",
+     "start_time": "2023-06-14T17:20:03.960665Z"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "def kwx_kws(text,top=None):\n",
+    "    res = []\n",
+    "    score = []\n",
+    "    candidates = keywords_selector(text)\n",
+    "    for c in candidates:\n",
+    "        if c not in history_score:\n",
+    "            res.append(c)\n",
+    "            score.append(1)\n",
+    "        elif history_score[c]>0:\n",
+    "            res.append(c)\n",
+    "            score.append(history_score[c])\n",
+    "        if c in global_kwx_index:\n",
+    "            for r in global_kwx_index[c]:\n",
+    "                if r not in candidates:\n",
+    "                    res.append(r)\n",
+    "                    score.append(global_kwx_index[c][r])\n",
+    "    if not res:\n",
+    "        res = ['None']\n",
+    "    sorted_res = [x for _, x in sorted(zip(score, res))][::-1]\n",
+    "    if top and len(sorted_res)>top:\n",
+    "        return sorted_res[:top]\n",
+    "    else:\n",
+    "        return sorted_res"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "86f377c6",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2023-06-14T17:17:18.438467Z",
+     "start_time": "2023-06-14T17:16:21.507730Z"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "evaluate_extractor(kwx, kwx_kws)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "cdfec2e8",
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2023-06-14T17:20:24.682626Z",
+     "start_time": "2023-06-14T17:20:06.517802Z"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "evaluate_extractor(kwx, kwx_kws, 10)"
+   ]
+  }
+ ],
+ "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": "3.9.12"
+  },
+  "toc": {
+   "base_numbering": 1,
+   "nav_menu": {},
+   "number_sections": true,
+   "sideBar": true,
+   "skip_h1_title": false,
+   "title_cell": "Table of Contents",
+   "title_sidebar": "Contents",
+   "toc_cell": false,
+   "toc_position": {},
+   "toc_section_display": true,
+   "toc_window_display": false
+  },
+  "varInspector": {
+   "cols": {
+    "lenName": 16,
+    "lenType": 16,
+    "lenVar": 40
+   },
+   "kernels_config": {
+    "python": {
+     "delete_cmd_postfix": "",
+     "delete_cmd_prefix": "del ",
+     "library": "var_list.py",
+     "varRefreshCmd": "print(var_dic_list())"
+    },
+    "r": {
+     "delete_cmd_postfix": ") ",
+     "delete_cmd_prefix": "rm(",
+     "library": "var_list.r",
+     "varRefreshCmd": "cat(var_dic_list()) "
+    }
+   },
+   "types_to_exclude": [
+    "module",
+    "function",
+    "builtin_function_or_method",
+    "instance",
+    "_Feature"
+   ],
+   "window_display": false
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}