cluster-analysis-gensim-tsne.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Cluster Ask.Cyberinfrastructure Posts with Gensim"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We are going to use Gensim, a Python library for extraction of semantic topics from documents to cluster the extracted posts from AskCI. \n",
    "\n",
    "We will start with raw text from posts, generate embeddings for each using Doc2Vec, and then cluster the embeddings to find the semantic structure of the posts using TSNE. The algorithms we will be using are unsupervised, meaning that we don't need to provide any labels for training.\n",
    "\n",
    "See the documentation at: <a>https://radimrehurek.com/gensim/index.html</a>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "from gensim.models.doc2vec import TaggedDocument, Doc2Vec\n",
    "from sklearn.manifold import TSNE\n",
    "import seaborn\n",
    "import matplotlib.pylab as plt\n",
    "%matplotlib inline\n",
    "\n",
    "import json\n",
    "import os\n",
    "import re\n",
    "\n",
    "date_folder = \"2019-10-19\"\n",
    "board = \"ask.cyberinfrastructure.org\"\n",
    "categories = [\"q-a\", \"Area-for-open-ended-topics-of-interest\"]\n",
    "\n",
    "topics = []\n",
    "labels = []\n",
    "meta = {}\n",
    "\n",
    "def read_json(json_file):\n",
    "    with open(json_file, 'r') as filey:\n",
    "        content = json.loads(filey.read())\n",
    "    return content\n",
    "        \n",
    "# Read in list of topics, each line compiled posts from it\n",
    "for category in categories:\n",
    "    json_path = os.path.join(date_folder, \"%s-%s-posts-%s.json\" %(board, category, date_folder))\n",
    "    content = read_json(json_path)\n",
    "\n",
    "    for label, sentences in content.items():\n",
    "        topics.append(' '.join(sentences))\n",
    "        labels.append(label)\n",
    "        \n",
    "        \n",
    "# Add metadata for each topic\n",
    "for category in categories:\n",
    "\n",
    "    # Metadata includes tags, likes, etc.\n",
    "    json_path = os.path.join(date_folder, \"%s-%s-topics-%s.json\" %(board, category, date_folder))\n",
    "    content = read_json(json_path)\n",
    "\n",
    "    for details in content:\n",
    "        meta[str(details['id'])] = details"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "# remove html formatting replace newlines\n",
    "def cleanhtml(raw_html):\n",
    "    cleanr = re.compile('<.*?>')\n",
    "    cleantext = re.sub(cleanr, '', raw_html)\n",
    "    removelinks = re.compile('http[s]?://(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\\(\\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+')\n",
    "    cleantext = re.sub(removelinks, '', cleantext)\n",
    "    cleantext = cleantext.replace('\\n', ' ')\n",
    "    cleantext = re.sub('\\d+', '', cleantext) # random numbers\n",
    "    cleantext = re.sub(\"[^a-zA-Z0-9 -]\", ' ', cleantext)\n",
    "    return cleantext\n",
    "\n",
    "cleaned = [cleanhtml(document) for document in topics]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# remove common words and tokenize\n",
    "stoplist = set('for a of the and to in'.split())\n",
    "texts = [[word for word in document.lower().split() if word not in stoplist]\n",
    "         for document in cleaned]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# remove words that appear only once\n",
    "from collections import defaultdict\n",
    "frequency = defaultdict(int)\n",
    "for text in texts:\n",
    "    for token in text:\n",
    "        frequency[token] += 1\n",
    "\n",
    "texts = [[token for token in text if frequency[token] > 1]\n",
    "         for text in texts]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Doc2Vec(dm/m,d5,n5,w2,s0.001,t4)\n",
      "179\n"
     ]
    }
   ],
   "source": [
    "# Generate tagged documents\n",
    "docs = []\n",
    "for i, text in enumerate(texts):\n",
    "    docs.append(TaggedDocument(text, [i]))\n",
    "\n",
    "# build model, wv is mapping between words and embeddings\n",
    "model = Doc2Vec(docs, vector_size=5, window=2, min_count=1, workers=4)\n",
    "\n",
    "# document embeddings\n",
    "print(model)\n",
    "print(len(model.docvecs.vectors_docs))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(179, 2)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7fbf8964a048>"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fbf8820fe80>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Try TSNE\n",
    "X_embedded = TSNE(n_components=2).fit_transform(model.docvecs.vectors_docs)\n",
    "print(X_embedded.shape)\n",
    "import pandas\n",
    "\n",
    "df = pandas.DataFrame(X_embedded)\n",
    "df.columns = ['dim1', 'dim2']\n",
    "\n",
    "plt.figure(figsize=(16,10))\n",
    "seaborn.scatterplot(\n",
    "    x='dim1',\n",
    "    y='dim2',\n",
    "    data=df,\n",
    "    palette=seaborn.color_palette(\"hls\", 10),\n",
    "    legend=\"full\",\n",
    "    alpha=0.3\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "    \"x\": 10.561829,\n",
      "    \"y\": -15.092673,\n",
      "    \"post_id\": \"303\",\n",
      "    \"posts_count\": 1,\n",
      "    \"tags\": [],\n",
      "    \"url\": \"https://ask.cyberinfrastructure.org/t/about-the-q-a-category/303/\",\n",
      "    \"views\": 125,\n",
      "    \"like_count\": 0,\n",
      "    \"title\": \"About the Q&A category\"\n",
      "}\n"
     ]
    }
   ],
   "source": [
    "# Notebooks are crap, save data to make a real plot in d3!\n",
    "# list of dicts with x,y, and link\n",
    "\n",
    "data = []\n",
    "for i, row in enumerate(X_embedded):\n",
    "    label = labels[i]\n",
    "    details = meta[label]\n",
    "    url = \"https://ask.cyberinfrastructure.org/t/%s/%s/\" %(details['slug'], label)\n",
    "    data.append({\"x\": round(row[0].astype(float), 6),  # from pandas\n",
    "                 \"y\": round(row[1].astype(float), 6),  # try to maintain original precision\n",
    "                 \"post_id\": label, \n",
    "                 \"posts_count\": details[\"posts_count\"], \n",
    "                 \"tags\": details[\"tags\"],\n",
    "                 \"url\": url,\n",
    "                 \"views\": details[\"views\"],\n",
    "                 \"like_count\": details[\"like_count\"],\n",
    "                 \"title\": details[\"title\"]})\n",
    "    \n",
    "print(json.dumps(data[0], indent=4))\n",
    "\n",
    "with open(\"askci-post-tsne-179x2.json\", 'w') as filey:\n",
    "    filey.writelines(json.dumps(data, indent=4))"
   ]
  },
  {
   "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}