ger_nlp_test.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Test German NLP Pipeline\n",
    "\n",
    "Test NLP pipeline for dependency parsing and semantic role labeling in German. \n",
    "\n",
    "Requirements:\n",
    "- `spacy`: >= 3.0\n",
    "- `pandas`: >= 1.0\n",
    "- `gensim`: >= 4.0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import spacy\n",
    "import re\n",
    "import math\n",
    "import warnings\n",
    "import gensim\n",
    "import random\n",
    "import numpy as np\n",
    "import seaborn as sns\n",
    "import pandas as pd\n",
    "from itertools import combinations\n",
    "from gensim import corpora\n",
    "from gensim.models.ldamodel import LdaModel\n",
    "from gensim.models.coherencemodel import CoherenceModel\n",
    "from urllib.request import urlopen\n",
    "from bs4 import BeautifulSoup"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "if spacy.__version__ < \"3\":\n",
    "    warnings.warn(\"Module 'spacy' should be version >= 3.0 to run this notebook without errors\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "if pd.__version__ < \"1.0\":\n",
    "    warnings.warn(\"Module 'pandas' should be version >= 1.0 to run this notebook without errors\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "if gensim.__version__ < \"4.0\":\n",
    "    warnings.warn(\"Module 'gensim' should be version >= 4.0 to run this notebook without errors\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Retrieve Stories From the Web\n",
    "\n",
    "Build web scraper for German stories about the pandemic. The stories are published on a webpage from the HSPV NRW (University of Applied Sciences for the police and public administration) and written by students as well as staff. For a general description, see: https://www.hspv.nrw.de/services/corona-krise/corona-geschichten"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define URL and story page ids\n",
    "\n",
    "URL = \"https://www.hspv.nrw.de/nachrichten/artikel/corona-geschichten\"\n",
    "\n",
    "ID_STORIES = [\"01\", \"02\", \"03\", \"04\", \"05\", \"06\", \"07\", \"08\", \"09\", \"10\", \"11\", \"12\", \"13\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to retrieve text from webpages\n",
    "\n",
    "def get_text_from_web(url, ids):\n",
    "    docs = []\n",
    "\n",
    "    for id in ids:\n",
    "        new_page = urlopen(url + \"-\" + id).read()\n",
    "\n",
    "        new_soup = BeautifulSoup(new_page, 'html.parser')\n",
    "\n",
    "        for script in new_soup([\"script\", \"style\"]):\n",
    "            script.decompose()\n",
    "        \n",
    "        docs.append(list(new_soup.stripped_strings))\n",
    "    \n",
    "    return docs\n",
    "\n",
    "  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Extract the relevant stories by removing text that is present in all documents (e.g., menu headers)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to extract only text from stories\n",
    "\n",
    "def extract_story_text(docs):\n",
    "    new_docs = [[s for s in docs[0] if s not in docs[-1]]]\n",
    "\n",
    "    for doc in docs[1:]:\n",
    "        new_docs.append([s for s in doc if s not in docs[0]])\n",
    "    \n",
    "    return new_docs"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Format the stories into a dictionary with keys:\n",
    "- title: Title\n",
    "- author: Author\n",
    "- date: Date\n",
    "- text: Text as single string\n",
    "\n",
    "Also removes dash unicode and hashtags in the text."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to format stories into dicts\n",
    "\n",
    "def convert_stories_to_dict(docs):\n",
    "    docs_dict = []\n",
    "    \n",
    "    for doc in docs:\n",
    "        new_dict = {}\n",
    "        title = doc[1]\n",
    "        author = doc[3]\n",
    "        date = doc[2]\n",
    "        text = \"\".join(s + \" \" for s in doc[1:] if s not in [title, author, date]).replace(\"\\xa0\", \"\")\n",
    "        docs_dict.append({\n",
    "            \"title\": title,\n",
    "            \"author\": author,\n",
    "            \"date\": date,\n",
    "            \"text\": text\n",
    "        })\n",
    "        \n",
    "    return docs_dict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "docs = get_text_from_web(URL, ID_STORIES)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "short_docs = extract_story_text(docs)\n",
    "\n",
    "dict_docs = convert_stories_to_dict(short_docs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Studienalltag in Zeiten von Corona Ann-Katrin Vengels, Studentin der HSPV NRW am Studienort Mülheim an der Ruhr, berichtet aus ihrem Alltag zwischen Homeschooling, Kinderbetreuung und Online-Studium. In den vergangenen Jahren habe ich die Frage nach meinem Befinden stets mit den Worten „Der normale Wahnsinn halt“ beantwortet. Und ja, irgendwie ist der Wahnsinn tatsächlich normal geworden. Ich bin Mutter von zwei Kindern, meine Tochter ist 13 und mein Sohn sechs. Seit zehn Jahren bin ich verheiratet. 2019 habe ich mich dazu entschlossen, meinen Beruf aufzugeben und mich weiterzuentwickeln, weshalb ich mich für ein duales Bachelorstudium an der HSPV entschieden habe. Mittlerweile studiere ich im zweiten Studienjahr Polizeivollzugsdienst und bin Kommissaranwärterin. Mein Ehemann (der inzwischen auch mein Kollege ist) und ich sind über die Jahre zu einem eingespielten Team geworden, das perfekt aufeinander eingestellt ist. Ein großer Pluspunkt, der definitiv eine wichtige Rolle bei meinem Berufswechsel gespielt hat. Als die Kinder schließlich groß genug waren und wir unser Haus gekauft hatten, sprach nichts mehr dagegen, sich beruflich neu zu orientieren – also wagte ich den Schritt. Der Studienplan war natürlich von Vorteil, denn der Verlauf der nächsten drei Jahre war absehbar und planbar. Mit der zugesicherten Unterstützung der Großeltern und der Familie konnte eigentlich nichts mehr schiefgehen. Doch dann kamen die vorgezogenen Osterferien im vergangen Frühjahr. Die Einschulung im Sommer schien zu diesem Zeitpunkt noch gar nicht in Gefahr, denn alles wirkte irgendwie unwirklich und noch nicht besorgniserregend. Doch dann überschlugen sich die Ereignisse: Die Einschulung mit Abstand und Singverbot, der erste Corona-Fall in der Familie, der Vater, den man nicht auf der Intensivstation besuchen konnte, Geburtstage und Weihnachten im engsten Familienkreis – es kam mir vor wie ein Dauerlauf ohne wirkliches Ziel. Und nun haben wir bereits Februar 2021. Seit gut sechs Wochen sind wir im Lockdown und damit auch im Homeschooling: Grundschule ersteKlasse vs. Gymnasium achteKlasse – und nicht zu vergessen der tägliche Kampf um das WLAN. Das Leben hat sich grundsätzlich verändert. Alles wurde umstrukturiert, weshalb viele unserer eingespielten Rituale nicht mehr umgesetzt werden können. Stattdessen werden die Nächte immer kürzer, denn Ruhe zum Arbeiten findet man oft nur noch in der Nacht. Manche Themen lassen auch keine andere Wahl. Die Seminararbeit über Kindeswohlgefährdung zu schreiben, ist zum Beispiel schwer, wenn das eigene Kind gerade mit Begeisterung die Welt der Buchstaben kennenlernt. Oft sind es auch kleine Dinge während der Vorlesungen, die einen aus der Konzentration reißen und die Rolle der Mutter erfordern. Sich anschließend wieder in die laufende Vorlesung einzufinden, ist nicht immer leicht. An dieser Stelle möchte ich mich gerne bei meinen Dozenten bedanken, denn mir wurde nicht nur viel Verständnis entgegengebracht, durch tolle Gespräche konnte ich auch viele private Einblicke erhalten. Lerngruppen haben sich während dieser Zeit ebenfalls gebildet, denn es wurde schnell klar, dass sich die Motivation gemeinsam schneller finden lässt. Insgesamt gewinnt die Online- Lehre immer mehr an Qualität und das Gefühl der Vertrautheit kehrt langsam zurück. Immer wieder gibt es Momente, in denen man merkt: Es gibt sie noch, die schöne Seite der Medaille. Unter anderem hat sich still und heimlich ein ganz neues Familienleben gebildet: Meine Tochter und ich haben zum Beispiel das Häkeln für uns entdeckt und hantieren gerne gemeinsam in der Küche – Sushi steht aktuell hoch im Kurs. Außerdem ist es toll, sich von einer 13-Jährigen Vorträge über Lernstrategien anzuhören. Immer häufiger teilen wir unsere Sorgen und ich kann mir einiges von ihr abschauen. Die täglichen Lernerfolge so nah erleben zu können und unmittelbar mitzubekommen, wie zum Beispiel Worte plötzlich einen Sinn ergeben, ist einfach toll und magisch zugleich. Im Großen und Ganzen überwiegt die Dankbarkeit – Dankbarkeit für die Gesundheit und für die Chancen, die wir haben. \n"
     ]
    }
   ],
   "source": [
    "print(dict_docs[1][\"text\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Dependency Parsing and Named Entity Recognition\n",
    "\n",
    "Before running the pipeline, the pipeline must be installed, e.g., via:\n",
    "\n",
    "`python3 -m spacy download de_core_news_sm`\n",
    "\n",
    "For details on the pipeline, see: https://spacy.io/models/de.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load german nlp pipeline trained on news documents (sm = small size)\n",
    "\n",
    "nlp = spacy.load(\"de_core_news_sm\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def analyse_stories(docs):\n",
    "    processed_docs = []\n",
    "    \n",
    "    for doc in docs:\n",
    "        processed_docs.append(nlp(doc[\"text\"]))\n",
    "        \n",
    "    return processed_docs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "processed_docs = analyse_stories(dict_docs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# From danlp_test notebook\n",
    "\n",
    "def print_results(doc):\n",
    "    analysis = []\n",
    "    for token in doc:\n",
    "        analysis.append({ \n",
    "            \"text\": token.text,\n",
    "            \"lemma\": token.lemma_,\n",
    "            \"head\": token.head.i,\n",
    "            \"dep\": token.dep_,\n",
    "            \"pos\": token.pos_,\n",
    "            \"morph_case\": \"\".join(token.morph.get(\"Case\")),\n",
    "            \"morph_person\": \"\".join(token.morph.get(\"Person\")),\n",
    "            \"morph_number\": \"\".join(token.morph.get(\"Number\")),\n",
    "            \"ent\": token.ent_type_\n",
    "        })\n",
    "    return pd.DataFrame(analysis)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# From danlp_test notebook\n",
    "\n",
    "def show_results(results_dict):\n",
    "    results_df = pd.DataFrame([{ \"key\": key, \"count\": results_dict[key]} \n",
    "                               for key in sorted(results_dict, key=lambda k: results_dict[k], reverse=True)])\n",
    "    return results_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "# From danlp_test notebook\n",
    "\n",
    "def get_best_keys(results_df):\n",
    "    best_count = -1\n",
    "    best_keys = []\n",
    "    for i, row in results_df.iterrows():\n",
    "        if row[\"count\"] > best_count:\n",
    "            best_count = row[\"count\"]\n",
    "            best_keys = [row[\"key\"]]\n",
    "        elif row[\"count\"] == best_count:\n",
    "            best_keys.append(row[\"key\"])\n",
    "    case_is_upper = []\n",
    "    for key in best_keys:\n",
    "        case_is_upper.append(re.search(r\"^[A-Z]\", key) != None)\n",
    "    if True in case_is_upper:\n",
    "        best_keys = [ best_keys[i] for i in range(0, len(best_keys)) if case_is_upper[i] ] \n",
    "    return best_keys"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "# From danlp_test notebook\n",
    "\n",
    "def get_actors(nlp_table_df):\n",
    "    actors = {}\n",
    "    for i, row in nlp_table_df.iterrows():\n",
    "        if row[\"dep\"] == \"sb\" and re.search(\"[a-zA-Z]\", row[\"text\"]):\n",
    "            actor = row[\"text\"]\n",
    "        else:\n",
    "            actor = \"\"\n",
    "        if actor != \"\":\n",
    "            if actor in actors:\n",
    "                actors[actor] += 1\n",
    "            else:\n",
    "                actors[actor] = 1\n",
    "    return actors"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# From danlp_test notebook\n",
    "\n",
    "def get_locations(nlp_table_df):\n",
    "    locations = {}\n",
    "    for i, row in nlp_table_df.iterrows():\n",
    "        if row[\"ent\"] == \"LOC\" and re.search(\"[a-zA-Z]\", row[\"text\"]):\n",
    "            location = row[\"text\"]\n",
    "        else:\n",
    "            location = \"\"\n",
    "        if location != \"\":\n",
    "            if location in locations:\n",
    "                locations[location] += 1\n",
    "            else:\n",
    "                locations[location] = 1\n",
    "    return locations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "# From danlp_test notebook\n",
    "\n",
    "def get_actions(nlp_table_df):\n",
    "    actions = {}\n",
    "    for i, row in nlp_table_df.iterrows():\n",
    "        if row[\"pos\"] == \"VERB\" and re.search(\"[a-zA-Z]\", row[\"text\"]):\n",
    "            if row[\"lemma\"] in actions:\n",
    "                actions[row[\"lemma\"]] += 1\n",
    "            else:\n",
    "                actions[row[\"lemma\"]] = 1\n",
    "    return actions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Calculate the type of agency in the story as the average frequency of pronouns, nouns (and proper nouns) in nominative (agent) vs. dative/accusative (object) case."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calc_agency(nlp_table_df):\n",
    "    agency = (nlp_table_df[nlp_table_df.morph_case.ne(\"\") \n",
    "                           & nlp_table_df.pos.isin([\"PROPN\", \"NOUN\", \"PRON\"]) \n",
    "                           & nlp_table_df.morph_case.ne(\"Gen\")]\n",
    "        .assign(agent_object = lambda dataframe: np.where(dataframe[\"morph_case\"] == \"Nom\", \"agent\", \"object\"))\n",
    "        .groupby([\"pos\", \"morph_person\", \"morph_number\", \"agent_object\"], as_index = False)\n",
    "        .agg(count = (\"text\", \"count\"))\n",
    "    )\n",
    "    \n",
    "    return agency"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "def concatenate_results(docs, fun):\n",
    "    df_list = []\n",
    "    for i, doc in enumerate(docs):\n",
    "        df_list.append(show_results(fun(print_results(doc))).assign(id = i+1))\n",
    "    \n",
    "    return pd.concat(df_list, ignore_index = True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Sum the counts for concatenated data frames."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "def show_top_keys(count_df_list):\n",
    "    top_keys = (count_df_list\n",
    "          .groupby([\"key\"], as_index = False)\n",
    "          .agg(count = (\"count\", sum))\n",
    "          .sort_values(by = [\"count\"], ascending = False)\n",
    "          .head(10)  \n",
    "    )\n",
    "    \n",
    "    return top_keys"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "def concatenate_agency(docs):\n",
    "    df_list = []\n",
    "    for i, doc in enumerate(docs):\n",
    "        df_list.append(calc_agency(print_results(doc)).assign(id = i+1))\n",
    "    \n",
    "    return pd.concat(df_list)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Plot the agency for pronouns in a bar chart."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_pronoun_agency(agency_df):\n",
    "    plot_df = agency_df[agency_df.pos.eq(\"PRON\") & agency_df.morph_number.ne(\"\") & agency_df.morph_person.ne(\"\")]\n",
    "    \n",
    "    plot = sns.catplot(\n",
    "        data = plot_df,\n",
    "        x = \"morph_person\",\n",
    "        y = \"count\",\n",
    "        hue = \"agent_object\",\n",
    "        col =\"morph_number\",\n",
    "        kind = \"point\"\n",
    "    )\n",
    "    \n",
    "    print(plot)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "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>key</th>\n",
       "      <th>count</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>203</th>\n",
       "      <td>ich</td>\n",
       "      <td>110</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>225</th>\n",
       "      <td>wir</td>\n",
       "      <td>52</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>201</th>\n",
       "      <td>es</td>\n",
       "      <td>30</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>66</th>\n",
       "      <td>Ich</td>\n",
       "      <td>29</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>210</th>\n",
       "      <td>man</td>\n",
       "      <td>19</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>216</th>\n",
       "      <td>sie</td>\n",
       "      <td>18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>190</th>\n",
       "      <td>das</td>\n",
       "      <td>17</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>192</th>\n",
       "      <td>die</td>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>222</th>\n",
       "      <td>was</td>\n",
       "      <td>13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Alltag</td>\n",
       "      <td>11</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        key  count\n",
       "203     ich    110\n",
       "225     wir     52\n",
       "201      es     30\n",
       "66      Ich     29\n",
       "210     man     19\n",
       "216     sie     18\n",
       "190     das     17\n",
       "192     die     16\n",
       "222     was     13\n",
       "4    Alltag     11"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Show top actors from all stories\n",
    "\n",
    "show_top_keys(concatenate_results(processed_docs, get_actors))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The storytellers themselves seem to be the most prevalent actors (\"ich\"). Quite often they also act as part of a group (i.e., \"wir\")."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "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>key</th>\n",
       "      <th>count</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>174</th>\n",
       "      <td>kommen</td>\n",
       "      <td>21</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>149</th>\n",
       "      <td>haben</td>\n",
       "      <td>17</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>135</th>\n",
       "      <td>gehen</td>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>53</th>\n",
       "      <td>berichten</td>\n",
       "      <td>15</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>132</th>\n",
       "      <td>geben</td>\n",
       "      <td>14</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>123</th>\n",
       "      <td>finden</td>\n",
       "      <td>14</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>189</th>\n",
       "      <td>machen</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>180</th>\n",
       "      <td>lassen</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>313</th>\n",
       "      <td>verändern</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>241</th>\n",
       "      <td>sehen</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           key  count\n",
       "174     kommen     21\n",
       "149      haben     17\n",
       "135      gehen     16\n",
       "53   berichten     15\n",
       "132      geben     14\n",
       "123     finden     14\n",
       "189     machen     10\n",
       "180     lassen     10\n",
       "313  verändern     10\n",
       "241      sehen     10"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Show top actions from all stories\n",
    "\n",
    "show_top_keys(concatenate_results(processed_docs, get_actions))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Interestingly, 'changing' (\"verändern\") seems to be quite a prevalent action."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "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>key</th>\n",
       "      <th>count</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>26</th>\n",
       "      <td>Homeoffice</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>52</th>\n",
       "      <td>Mülheim</td>\n",
       "      <td>6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>63</th>\n",
       "      <td>Ruhr</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>84</th>\n",
       "      <td>der</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>83</th>\n",
       "      <td>an</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>41</th>\n",
       "      <td>Lehrveranstaltungen</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>44</th>\n",
       "      <td>Lockdown</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>Borken</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>35</th>\n",
       "      <td>Kreis</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>Corona</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                    key  count\n",
       "26           Homeoffice      9\n",
       "52              Mülheim      6\n",
       "63                 Ruhr      4\n",
       "84                  der      4\n",
       "83                   an      4\n",
       "41  Lehrveranstaltungen      4\n",
       "44             Lockdown      3\n",
       "9                Borken      3\n",
       "35                Kreis      3\n",
       "14               Corona      3"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Show top locations from all stories\n",
    "\n",
    "show_top_keys(concatenate_results(processed_docs, get_locations))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Some of the keys are not actually locations (i.e., lockdown, Corona). Apparently, spaCy's named entity recognition is not very good for German due to limited training data (e.g., see https://github.com/explosion/spaCy/issues/2774). However, \"Homeoffice\", \"Mülheim\", and \"Lehrveranstaltungen\" (i.e., in class) seem like very plausible locations."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<seaborn.axisgrid.FacetGrid object at 0x000002581165FD30>\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 792.5x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Show agency across all stories\n",
    "\n",
    "plot_pronoun_agency(concatenate_agency(processed_docs))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Semantic Role Labeling\n",
    "\n",
    "Only option sofar: With pre-trained tensorflow model as described in https://liir.cs.kuleuven.be/software_pages/damesrl.php. The link also contains the source code and pre-trained model files in folder `charatt_ger_march23`. The software is distributed under the Apache 2.0 license."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Topic Modeling\n",
    "\n",
    "Define a corpus for topic modeling that only includes lemmatized adjectives, nouns, proper nouns, and verbs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to create corpus for topic modeling\n",
    "\n",
    "def create_corpus_from_processed_docs(docs):\n",
    "    corpus = []\n",
    "\n",
    "    for doc in docs:\n",
    "        new_res = print_results(doc)\n",
    "        new_lemma_doc = new_res.lemma[new_res.pos.isin([\"ADJ\", \"NOUN\", \"PNOUN\", \"VERB\"])]\n",
    "        corpus.append(new_lemma_doc)\n",
    "\n",
    "    return corpus"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create corpus\n",
    "\n",
    "corpus = create_corpus_from_processed_docs(processed_docs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Latent Dirichlet Allocation\n",
    "\n",
    "One method for topic modeling is Latent Dirichlet Allocation (LDA; see Blei, Ng, & Jordan, 2003). In LDA, each document in a corpus is modeled as a finite mixture of topics (i.e., the number of topics is known in advance and fixed). Each topic is modeled as an infinite mixture of words (i.e., the number of words per topic is not known in advance and random). The argument `passes` specifies how often the entire corpus is passed through during training."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to apply LDA to corpus\n",
    "\n",
    "def apply_LDA(corpus, num_topics, seed = None):\n",
    "    dictionary = corpora.Dictionary(corpus)\n",
    "    \n",
    "    doc_term_matrix = [dictionary.doc2bow(doc) for doc in corpus]\n",
    "    \n",
    "    lda_fit = LdaModel(doc_term_matrix, num_topics = NUM_TOPICS, id2word = dictionary, passes = 100, random_state = seed)\n",
    "    \n",
    "    return dictionary, doc_term_matrix, lda_fit"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Apply LDA to corpus\n",
    "\n",
    "NUM_TOPICS = 3\n",
    "\n",
    "SEED = 791234\n",
    "\n",
    "dictionary, doc_term_matrix, lda_fit = apply_LDA(corpus, NUM_TOPICS, SEED)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Funtion to get top word probabilities per topic as data frame\n",
    "\n",
    "def get_top_word_probs(lda_fit, num_topics, num_words):\n",
    "    top_words = lda_fit.show_topics(num_topics = num_topics, num_words = num_words, formatted = False)\n",
    "\n",
    "    top_words_df = (pd.DataFrame(top_words)\n",
    "                      .explode(1, ignore_index = True)\n",
    "                      .rename(columns = {0: \"topic_id\", 1: \"word_prob\"})\n",
    "                     )\n",
    "    top_words_df[[\"word\", \"prob\"]] = top_words_df[\"word_prob\"].tolist()\n",
    "    \n",
    "    return top_words_df[[\"topic_id\", \"word\", \"prob\"]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to plot top LDA word probabilities per topic\n",
    "\n",
    "def plot_top_word_probs(lda_fit, num_topics, num_words):\n",
    "    plot_df = get_top_word_probs(lda_fit, num_topics, num_words)\n",
    "    \n",
    "    plot = sns.catplot(\n",
    "        data = plot_df,\n",
    "        x = \"prob\",\n",
    "        y = \"word\",\n",
    "        col = \"topic_id\",\n",
    "        hue = \"topic_id\",\n",
    "        kind = \"bar\",\n",
    "        sharey = False,\n",
    "        col_wrap = 3\n",
    "    )\n",
    "    \n",
    "    return plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<seaborn.axisgrid.FacetGrid at 0x25816970730>"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_top_word_probs(lda_fit, NUM_TOPICS, 5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Term Score\n",
    "According to Blei & Lafferty (2009), it is often more useful to look at the term-score metric for each topic, calculated as follows (p. 75): $$\\text{term-score}_{wi} = \\beta_{wi}\\log\\left(\\frac{\\beta_{wi}}{(\\prod_{j = 1}^{K}\\beta_{wj})^{1/K}}\\right),$$ for each word $w$ with probability $\\beta_{wi}$ to to appear in topic $i$ with $K$\n",
    " being the number of topics in the model. This score is similar to the tf/idf metric and takes the probability of a word appearing in the other topics into account."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Funtion to get top word scores per topic as data frame\n",
    "\n",
    "def get_top_word_scores(lda_fit, num_topics, num_words):\n",
    "    \n",
    "    word_probs = lda_fit.get_topics()\n",
    "    word_token = [dictionary.id2token[i] for i in range(len(dictionary))]\n",
    "    word_probs_df = pd.DataFrame(word_probs).transpose()\n",
    "    word_probs_df[\"prod\"] = word_probs_df.prod(axis=1)\n",
    "    word_probs_df.insert(0, \"word\", word_token)\n",
    "    word_probs_df = word_probs_df.melt(id_vars = [\"word\", \"prod\"], var_name = \"topic_id\", value_name = \"prob\")\n",
    "    word_probs_df = (word_probs_df\n",
    "                    .assign(score = word_probs_df[\"prob\"]*np.log(word_probs_df[\"prob\"]/(word_probs_df[\"prod\"])**(1/num_topics)))\n",
    "                    .sort_values(by = [\"topic_id\", \"score\"], ascending = False)\n",
    "                    .groupby(\"topic_id\", as_index = False)\n",
    "                    .head(num_words)\n",
    "                    )\n",
    "\n",
    "    return word_probs_df[[\"topic_id\", \"word\", \"score\"]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to plot top word scores per topic\n",
    "\n",
    "def plot_top_word_scores(lda_fit, num_topics, num_words):\n",
    "    plot_df = get_top_word_scores(lda_fit, num_topics, num_words)\n",
    "    \n",
    "    plot = sns.catplot(\n",
    "        data = plot_df,\n",
    "        x = \"score\",\n",
    "        y = \"word\",\n",
    "        col = \"topic_id\",\n",
    "        hue = \"topic_id\",\n",
    "        kind = \"bar\",\n",
    "        sharey = False,\n",
    "        col_wrap = 3\n",
    "    )\n",
    "    \n",
    "    return plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<seaborn.axisgrid.FacetGrid at 0x2583f409c10>"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot top word scores for each topic\n",
    "\n",
    "plot_top_word_scores(lda_fit, NUM_TOPICS, 5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Topic differences\n",
    "To get insight into the differences between topics, the log (base 2) of the probability ratios between two topics will be calculated for each word. For each word $w$ and topics $i$ and $j$: $$ D_{wij} = log_2\\left(\\frac{\\beta_{wi}}{\\beta_{wi}}\\right),$$ where $\\beta_{wi}$ is the probability that word $w$ occurs in topic $i$. For example, a $D_{wij} = 2$ would indicate that word $w$ is two times more likely to occur within topic $i$ than in topic $j$."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to calculate the differences between topics\n",
    "\n",
    "def get_topic_differences(lda_fit, num_topics, num_words):\n",
    "    word_probs = lda_fit.get_topics()\n",
    "    word_token = [dictionary.id2token[i] for i in range(len(dictionary))]\n",
    "    word_probs_df = pd.DataFrame(word_probs).transpose()\n",
    "    word_probs_df.insert(0, \"word\", word_token)\n",
    "    word_probs_df = word_probs_df[word_probs_df[0].ge(0.001) & word_probs_df[1].ge(0.001) & word_probs_df[2].ge(0.001)]\n",
    "    \n",
    "    for c in combinations(range(NUM_TOPICS), 2):\n",
    "        word_probs_df[\"{}-{}\".format(c[0], c[1])] = np.log2(word_probs_df[c[0]]/word_probs_df[c[1]])\n",
    "        \n",
    "    word_probs_df = (word_probs_df\n",
    "                    .drop([i for i in range(num_topics)], axis = 1)\n",
    "                    .melt(id_vars = \"word\", var_name = \"diff\", value_name = \"value\")\n",
    "                    )\n",
    "                    \n",
    "    word_probs_df = (word_probs_df\n",
    "                    .assign(abs_value = np.abs(word_probs_df[\"value\"]))\n",
    "                    .sort_values(by = [\"diff\", \"abs_value\"], ascending = False)\n",
    "                    .groupby(\"diff\", as_index = False)\n",
    "                    .head(num_words)\n",
    "                    )\n",
    "    \n",
    "    return word_probs_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to plot differences between topics\n",
    "\n",
    "def plot_topic_differences(lda_fit, num_topics, num_words):\n",
    "    plot_df = get_topic_differences(lda_fit, num_topics, num_words)\n",
    "    \n",
    "    plot = sns.catplot(\n",
    "        data = plot_df,\n",
    "        x = \"value\",\n",
    "        y = \"word\",\n",
    "        col = \"diff\",\n",
    "        kind = \"bar\",\n",
    "        sharey = False,\n",
    "        col_wrap = 3\n",
    "    )\n",
    "    \n",
    "    plot.set(xlabel = \"log2(prob_ratio)\")\n",
    "    \n",
    "    return plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\MalteLuken\\Repositories\\navigating-stories\\lib\\site-packages\\seaborn\\categorical.py:3808: UserWarning: Setting `sharey=False` with `color=None` may cause different levels of the `y` variable to share colors. This will change in a future version.\n",
      "  warnings.warn(msg.format(\"sharey\", \"y\"), UserWarning)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<seaborn.axisgrid.FacetGrid at 0x25839c335e0>"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot differences between topics\n",
    "\n",
    "plot_topic_differences(lda_fit, NUM_TOPICS, 10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Topic Probabilities per Document"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to get topic probabilities per document\n",
    "\n",
    "def get_topic_probs(lda_fit, doc_term_matrix):\n",
    "    doc_topic_probs = lda_fit.get_document_topics(doc_term_matrix, minimum_probability = 0.0)\n",
    "\n",
    "    doc_topic_probs_df = pd.DataFrame([[t[1] for t in l] for l in doc_topic_probs])\n",
    "    \n",
    "    return doc_topic_probs_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "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>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.002019</td>\n",
       "      <td>0.002086</td>\n",
       "      <td>0.995896</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.001725</td>\n",
       "      <td>0.996527</td>\n",
       "      <td>0.001748</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.003067</td>\n",
       "      <td>0.003061</td>\n",
       "      <td>0.993872</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.001391</td>\n",
       "      <td>0.997202</td>\n",
       "      <td>0.001407</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.002488</td>\n",
       "      <td>0.994875</td>\n",
       "      <td>0.002636</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>0.001372</td>\n",
       "      <td>0.001393</td>\n",
       "      <td>0.997234</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>0.992656</td>\n",
       "      <td>0.003570</td>\n",
       "      <td>0.003774</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>0.995208</td>\n",
       "      <td>0.002360</td>\n",
       "      <td>0.002432</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>0.001479</td>\n",
       "      <td>0.997015</td>\n",
       "      <td>0.001507</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>0.997654</td>\n",
       "      <td>0.001166</td>\n",
       "      <td>0.001180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0.002843</td>\n",
       "      <td>0.002803</td>\n",
       "      <td>0.994354</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>0.001941</td>\n",
       "      <td>0.001948</td>\n",
       "      <td>0.996111</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>0.002776</td>\n",
       "      <td>0.002795</td>\n",
       "      <td>0.994429</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           0         1         2\n",
       "0   0.002019  0.002086  0.995896\n",
       "1   0.001725  0.996527  0.001748\n",
       "2   0.003067  0.003061  0.993872\n",
       "3   0.001391  0.997202  0.001407\n",
       "4   0.002488  0.994875  0.002636\n",
       "5   0.001372  0.001393  0.997234\n",
       "6   0.992656  0.003570  0.003774\n",
       "7   0.995208  0.002360  0.002432\n",
       "8   0.001479  0.997015  0.001507\n",
       "9   0.997654  0.001166  0.001180\n",
       "10  0.002843  0.002803  0.994354\n",
       "11  0.001941  0.001948  0.996111\n",
       "12  0.002776  0.002795  0.994429"
      ]
     },
     "execution_count": 92,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Extract topic probabilities per document\n",
    "\n",
    "get_topic_probs(lda_fit, doc_term_matrix)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Document Similarity\n",
    "To inspect the similarity between documents, Blei & Lafferty (2009) suggest using the Heller distance between documents $f$ and $g$ as $$d_{fg} = \\sum_{k = 1}^{K}\\left(\\sqrt{\\gamma_{fk}} - \\sqrt{\\gamma_{gk}}\\right)^2, $$ where $\\gamma_{fk}$ is the probability of document $f$ for topic $k$ with $K$ being the total number of topics. Higher $d$ indicates less similarity between two documents. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to compute document similarity\n",
    "\n",
    "def get_doc_similarity(lda_fit, doc_term_matrix):\n",
    "    topic_probs = get_topic_probs(lda_fit, doc_term_matrix)\n",
    "    \n",
    "    doc_similarity = []\n",
    "    \n",
    "    for i in range(len(topic_probs)):\n",
    "        row = []\n",
    "        \n",
    "        for j in range(len(topic_probs)):\n",
    "            new_similarity = np.sum((np.sqrt(topic_probs.iloc[i,:]) - np.sqrt(topic_probs.iloc[j,:]))**2)\n",
    "            row.append(new_similarity)\n",
    "        \n",
    "        doc_similarity.append(row)\n",
    "            \n",
    "    return pd.DataFrame(doc_similarity)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to plot document similarity\n",
    "\n",
    "def plot_doc_similarity(lda_fit, doc_term_matrix):\n",
    "    plot_df = get_doc_similarity(lda_fit, doc_term_matrix)\n",
    "    \n",
    "    sns.set_theme()\n",
    "    \n",
    "    plot = sns.heatmap(plot_df)\n",
    "    \n",
    "    return plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:>"
      ]
     },
     "execution_count": 113,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWgAAAD7CAYAAABHYA6MAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/YYfK9AAAACXBIWXMAAAsTAAALEwEAmpwYAAApH0lEQVR4nO3de1hU9b4/8PcIM1z0h6Yxk2InttnFFJTAFPOSJiAKEhfLSwGSpBVgnJNbtmEQW9LMLZpYmW07mOydFDocejjoJk49KnRKnyfFvJy2bky8gJdquI0Mw/r94dMUgnNh1gyL5fvls56HtdbMZ31H4OPX7/quz1chCIIAIiKSnH693QAiIuoeEzQRkUQxQRMRSRQTNBGRRDFBExFJFBM0EZFEuTr1YiofUeO11O4XNR4AePqGih6z9VyFqPGEtlZR4wGA5wORosZrObVX1HgAAKWb6CEVripxA/YTv8/jce8M0WP2Be1tF+x6v+HqWatfq7x7hF3XchSnJmgiIqfpMPZ2C+zGBE1E8mRs7+0W2I0JmohkSRA6ersJdmOCJiJ56mCCJiKSJvagiYgk6k64SXjmzBns27cPly9fRr9+/aBWqzFlyhT4+fk5o31ERD0jgx602UmbhYWF+Pd//3cAgJ+fH0aPHg0AWL16NXbs2OH41hER9ZBgbLd6s1VTUxMiIiJQV1fX6fjJkycRFRVl2qZMmYKIiAgAgFarxeTJk03n8vLyLF7HbA96586d0Gq18PDw6HR88eLFiI6ORlJSkq2fi4jIORx0k/Do0aPIzMxEbW1tl3OjRo1CSUkJAKC1tRXz5s1DdnY2AKCmpgYZGRmmhG0Nsz1oV1dXtLd3/ddFr9dDqVRafREiIqcTOqzfbFBUVISsrCyo1Wqzr9u2bRvGjx+PoKAgADcTtFarxdy5c/Hqq6/il19+sXgtsz3oZcuW4amnnkJwcDC8vb0BAFeuXMHXX3+N9PR0az8PEZHz2XCTUKfTQafTdTnu5eUFLy+vTsdyc3OtildUVITS0lLTMW9vb7zwwgvw9/fHxo0bkZOTg7/85S9m45hN0JGRkXjsscdQXV2NhoYGCIKAoKAgpKamQqPRWGwkEVGvsaFnXFBQgPz8/C7HU1JSkJqaavOlS0tLMXPmTAwZMsR0bOvWraavlyxZgpkzZ1qMY3EWh0ajwVNPPWVzA4mIepUNN/8SEhIQHR3d5fitvWdrVVRUYOnSpab9xsZGFBcXIzExEQAgCAJcXS3PcuY8aCKSJxtuEnY3lNFTgiDg+++/R0BAgOmYp6cnPvzwQwQEBGDs2LHYtWsXQkJCLMZiPWgikiVBMFq92Ss5ORk1NTUAgOvXr0OpVMLN7bfyuC4uLti0aROys7MRHh6O77//HitWrLAYVyEIgmB366zEetDiYD1o8bAetHTZWw9a/93nVr/WfZz1U9+cyalDHGInVEck05YfSi2/yEYe91m+GdDbWs9XihpPaG0UNR4AQOGA5Cfy96a17ktR4wFAy5kycQO6OODXXoqlPVksiYhIomTwqDcTNBHJk9HQ2y2wGxM0EckThziIiCSKQxxERBLFHjQRkUQxQRMRSZMg95uEFy9eNPvmYcOGidoYIiLRyH0MeunSpaitrYVarcatDxwqFAp88cUXDm0cEVGPyX2I4+9//zsWLlyIrKwsBAYGOqtNRET2k0EP2uyzswMGDMCaNWug1Wqd1BwiIpF0dFi/SZTFm4T+/v7w9/d3RluIiMQjgx40Z3EQkTx1s55qX8METUTyxB40EZFESXhs2VpOTdBi1292RO1msQvXA+LX83VEwX6xi8I3H/ubqPEAQKES/xdO/O+NXtR4AOB5/2zRY/YF9hbsZw+aiEiq2IMmIpIo9qCJiCSKsziIiCTKeethO4z4q3ASEUmBA58kbGpqQkREBOrq6rqcy8/Px/Tp0xEVFYWoqCgUFhYCAE6ePInY2FiEhYXhtddeQ7sVPXwmaCKSJwcl6KNHj2LBggWora3t9vzx48exceNGlJSUoKSkBIsWLQIArFixAqtXr8a+ffsgCAKKioosXstigq6oqMDHH3+MH3/8sdPx3bt3W/FRiIh6idBh9abT6VBXV9dl0+l0XcIWFRUhKysLarW628seP34c27dvR2RkJHJycnDjxg1cuHABer0e48aNAwDExMSgvLzc4kcwm6A3bNiAXbt2oba2FvPnz0dJSYnp3CeffGIxOBFRrzEard4KCgrw5JNPdtkKCgq6hM3NzUVQUFC3l2xubsaoUaOwcuVK7N27FzqdDu+++y4aGhrg7e1tep23tzfq6+stfgSzNwm/+uor7N27F66urnjuueeQlJQElUqF8PDwLvWhiYgkxYahi4SEBERHR3c57uXlZdMl+/fvj+3bt5v2k5KSsGrVKkybNq3LaxUKhcV4ZhO0IAimIL6+vti2bRsWL16MwYMHWxWciKjX2JCgvby8bE7G3bl48SKqqqoQFxcH4GYOdXV1hUajwdWrV02vu3Llym2HSH7P7BDHrFmz8Nxzz+HYsWMAgAceeACbN2/GK6+80mVMmohIUmwYgxaLu7s73n77bZw/fx6CIKCwsBAhISHw8fGBm5sbjhw5AgDQarWYOnWqxXhme9ApKSkIDAxE//79TccCAwOxZ88e7Nixw86PQkTkOEKH84Zhk5OTkZaWBj8/P+Tk5ODFF1+EwWDAo48+isWLFwO4eU8vMzMTzc3NeOSRRxAfH28xrkJw4mCyq8pH1HgsliSe/qNiRY3nmGJJHqLHhNJN3HgK8Weueo6YJXrMvsDeYkkt7y+3+rWeyzbbdS1H4ZOERCRPRmNvt8BuTNBEJE+sZkdEJFFM0LZpPVchajyP+2aKGg8Qf7wY6BsF11svHhA1ntD8s6jxAIg/XgzxF5EQ++8RAFrOWn7izBYKpUrUeAAgGNpEj2k3GTyrwR40EckTe9BERBLlxGl2jsIETUTyxFkcRETSJHCIg4hIojjEQUQkUXfCorG1tbXw8PCARqPBp59+itOnT+PRRx/F7NnSnzpGRHcwufeg//M//xMff/wxOjo6MHHiRFy6dAkhISEoLi7Gv/71L7z88svOaicRkW3aZX6TsLi4GGVlZbh69SoiIiLw9ddfw83NDfPmzUNcXBwTNBFJl9yHODo6OqBSqeDj44OkpCS4uf32JJdRBlNYiEjGZDDEYbY2YmhoKJ599lkYjUakpqYCAE6dOoWFCxciPDzcKQ0kIuoJoaPD6k2qzPagly9fjm+//RYuLi6mYyqVCqmpqd2usUVEJBky6EFbnMUxfvz4TvsjRozAiBEjHNYgIiJR3AkJmoioT5LBfTImaCKSJWeuSegoTNBEJE9M0LZxxGKnYusLbXQEsT+30KYXNR4AKESP6ACGG+LHFPt744j5wY743PZy4OyMpqYmzJ8/H++//z6GDx/e6VxFRQW2bNkCQRAwfPhwrF27FgMHDoRWq8WGDRswZMgQAMATTzyB9PR0s9dhD5qI5MlBPeijR48iMzMTtbW1Xc41NTUhOzsbxcXF0Gg02Lx5M7Zs2YLMzEzU1NQgIyMDERERVl9L/DXiiYikoEOwfrNBUVERsrKyoFaru5wzGAzIzs6GRqMBADz00EO4dOkSAKCmpgZarRZz587Fq6++il9++cXitZigiUiWBGOH1ZtOp0NdXV2XTafTdYmbm5uLoKCgbq951113YebMm2ul6vV6fPDBB6Z9b29vpKamoqSkBEOHDkVOTo7Fz8AhDiKSJxt6xgUFBcjPz+9yPCUlxfQUtS0aGxvx0ksv4eGHH0Z0dDQAYOvWrabzS5YsMSVuc5igiUiWbJlml5CQYEqkv+fl5WXzdRsaGvD8889j4sSJWLVqFYCbCbu4uBiJiYk32yYIcHW1nH6ZoIlInmxI0F5eXj1KxrcyGo1YtmwZwsPD8dJLL5mOe3p64sMPP0RAQADGjh2LXbt2ISQkxGI8mxL0unXrkJGRYXuriYiczYk1kJKTk5GWlobLly/jxIkTMBqN2LdvHwBgzJgxyM3NxaZNm5CdnQ29Xg9fX1+sX7/eYtzbJug//elPXY5VVlaa7jyuXbu2p5+FiMjhhHbHZujKykrT19u3bwcA+Pn54dSpU92+PigoCHv37rXpGrdN0IMGDYJWq8WyZctMXf+vv/4ajz32mE0XICLqFdKtImq1206zW7lyJTZu3IiysjIMGzYM0dHRGDhwIKKjo7sdTCcikhKhQ7B6kyqzY9DBwcEYNWoUsrKy8OWXX3IVFSLqO+Tcg/7VoEGDsHnzZowYMQLe3t7OaBMRkd1k34P+vXnz5mHevHmObAsRkXhk0IPmPGgikiWhvbdbYD8maCKSJUdUVXU2pyZozwciRY3Xer7S8ots5HHvDNFjtl48IGo8R9Ss9vQNFTVeyw+losYDACjdRA/ZWveluAH7uVh+jY0877Ncs0GO2tsu2BeACZqISJrYgyYikigmaCIiiRKMfWKRNLOYoIlIltiDJiKSKKGDPWgiIkmSfQ/62LFj8Pf3BwBUV1fjq6++gqurK0JCQjB27FinNJCIqCcEoe/3oM3W4sjKygIAFBYW4s0338Q999yDu+++G6+//jp27drllAYSEfWE0GH9JlVWDXEUFRVh586duOuuuwAAcXFxiIuLw7PPPuvQxhER9VSH3GdxtLe3o6OjA0OGDIGnp6fpuEqlQr9+FgvhERH1GjncJDSbZe+66y5MmzYN//znP03DHdXV1Zg/fz5mzZrllAYSEfWE0KGwepMqsz3onTt3AgDOnj0LnU4H4GbvOS0tDU888YTDG0dE1FOCdMs8W82qcYoRI0Zg3LhxAIDAwEAmZyKSPEf2oJuamhAREYG6urou506ePInY2FiEhYXhtddeQ3v7zbqnFy9exKJFizBr1iy8+OKLaG5utngdDiQTkSwJgsLqzRZHjx7FggULUFtb2+35FStWYPXq1di3bx8EQUBRUREA4I033sDChQtRXl6OMWPG4N1337V4LSZoIpIlo1Fh9WaLoqIiZGVlQa1Wdzl34cIF6PV604hDTEwMysvLYTAY8O233yIsLKzTcUv4JCERyZItPWOdTme6z/Z7Xl5e8PLy6nQsNzf3tnEaGho6rd3q7e2N+vp6/PTTTxgwYABcXV07HbfEqQm65dReUeMJrY2ixgOA5mN/Ez2m0PyzuPHa9KLGA8QvsC/24gyO0hc+d/Px3aLGU7iqRI0HAEJ7m+gx7WXL2HJBQQHy8/O7HE9JSUFqaqr11+zmzqRCobjtcUvYgyYiWbJlFkdCQgKio6O7HL+192yJRqPB1atXTftXrlyBWq3G4MGD0dTUBKPRCBcXF9NxS5igiUiWbOlBdzeU0RM+Pj5wc3PDkSNHEBgYCK1Wi6lTp0KpVCIoKAhlZWWIjIw0HbeENwmJSJaMHf2s3uyVnJyMmpoaAMCGDRuwdu1ahIeHo7W1FfHx8QBu1jYqKirC7NmzcfjwYbzyyisW4yqE7gZHHKTt7DfiBnTA4+aC3vLcRFspPP6fqPEcMQatULmLGo9j0OK5U8eg3R6cbNf7j/la/73wr3XAIsci4BAHEclShwzKjTJBE5Esyb4eNAAcOHDAND9Qq9UiJycHxcXFDm8YEZE9BMH6TarMJujc3Fxs27YNN27cwKZNm1BaWoqRI0fiH//4B9asWeOsNhIR2axDUFi9SZXZIY5Dhw6htLQULi4u+PLLL1FUVASVSoVnnnkGERERzmojEZHNxJid0dvMfgJ3d3dcu3YNADBkyBC0tLQAAFpbW02PLBIRSZFgwyZVZrNsSkoK4uLiMGfOHIwYMQLPPfccgoODcfDgQSxZssRZbSQispmUhy6sZTZBz5gxAw888AAqKipw7tw5jBs3Dv3798e6detMq30TEUmRHGZxWBynuPfee7F48WJntIWISDQSXqzbahxIJiJZEnAH9KCJiPqi9jthiIOIqC9iD9pWSjdx4ynEn+eoUDlg5Erkz+2QHzuxvzd9RR/43GIXsoKLUtx4ABQOKFxmL45BExFJFHvQREQSxR40EZFEGdmDJiKSJhtWvJIsJmgikqUO9qCJiKRJykWQrGV2bsyaNWvwyy+/OKstRESi6bBhkyqzCVqr1eLpp5/G/v37ndUeIiJRdCgUVm9SZTZBDx8+HFu3bsXOnTsxb948lJWVQa8Xf0VpIiKxGW3YbFFaWorZs2cjJCQEhYWFnc6dPHkSUVFRpm3KlCmmxU20Wi0mT55sOpeXl2fxWmbHoBUKBUaOHIldu3ahqqoKu3fvRm5uLnx9fXHPPffgL3/5i40fjYjIORwxi6O+vh55eXnYs2cPVCoV5s+fjwkTJmDkyJEAgFGjRqGkpATAzYVN5s2bh+zsbABATU0NMjIybFqNymyCFn63muKkSZMwadIkGAwGnD59GufPn7f1sxEROY0tszh0Op1pcezf8/LygpeXl2m/qqoKEydOxKBBgwAAYWFhKC8vR0pKSpf3btu2DePHj0dQUBCAmwn63Llz+OCDD/Dggw9i9erVGDhwoNl2mR3iWLRoUZdjSqUSY8aMQXh4uNnARES9yZYlrwoKCvDkk0922QoKCjrFbGhogLe3t2lfrVajvr6+y7V1Oh2Kioo6JW5vb2+kpqaipKQEQ4cORU5OjsXPYLYHPW/ePIsBiIikyJYhjoSEBERHR3c5/vveM9B5VOFXim5uMpaWlmLmzJkYMmSI6djWrVtNXy9ZsgQzZ8602C7OgyYiWbJl+tytQxm3o9FocPjwYdN+Q0MD1Gp1l9dVVFRg6dKlpv3GxkYUFxcjMTERwM1Eb83C29KrEUhEJAKjwvrNWpMmTUJ1dTWuX7+O1tZW7N+/H1OnTu30GkEQ8P333yMgIMB0zNPTEx9++CGOHj0KANi1axdCQkIsXo89aCKSJUc8gKLRaJCeno74+HgYDAbExcXB398fycnJSEtLg5+fH65fvw6lUgk3t99qjbu4uGDTpk3Izs6GXq+Hr68v1q9fb/F6CqG7QRUHMVw6KWo8j/ssj+HYquVMmegxPe+fLXpMsbXWfSlqPKG1UdR4ABxSXN/TN1TUeK0XD4gaDwAEfbOo8RQu4vfLBGO76DFVw/3sev+24c9a/dqldbvsupajsAdNRLIkgyUJmaCJSJ6kXGPDWkzQRCRLtj7CLUVM0EQkSyzYT0QkUXfEEEd1dTXc3d0REBCAHTt24JtvvsGYMWPwwgsvQKVSOaONREQ2k32CXr9+PQ4fPoz29nYMHz4cCoUCCxYsQGVlJXJycrBmzRpntZOIyCZyWFHFbII+cOAASkpK0NbWhmnTpuHgwYNQKpWYOnUqoqKinNVGIiKbyWEM2uyj3oIgoLGxET/99BP0ej2ampoAAHq9HgaDwSkNJCLqCUcV7Hcmsz3o5ORkhIaGQhAErFixAklJSQgODkZ1dTViY2Od1UYiIpt1yGCQw2yCjoqKQlhYGIxGI/r374/x48fj4MGDePXVV/H44487q41ERDaT/U1CAHB3dzd9/dBDD+Ghhx5yaIOIiMTQ9/vPnAdNRDJ1R/SgiYj6IjnM4mCCJiJZMspgkMO5CbqfuAu4iF3DGACENr3oMUWvEWy4IW48AOjnImo4zwciRY3nKGJ/bzyGTRE1HgC01O4XN6AD6kFDkN6AgvRaZDv2oIlIlmQ/zY6IqK/q++mZCZqIZIpDHEREEiWHm4Ti3rUjIpKIDhs2W5SWlmL27NkICQlBYWFhl/P5+fmYPn06oqKiEBUVZXrNyZMnERsbi7CwMLz22mtob7e80C570EQkS4IDetD19fXIy8vDnj17oFKpMH/+fEyYMAEjR440veb48ePYuHEjAgICOr13xYoVWLNmDcaNG4dVq1ahqKgICxcuNHs9iwm6oqICFRUVuHLlCpRKJf7t3/4N4eHhXS5ORCQljhiDrqqqwsSJEzFo0CAAQFhYGMrLy5GSkmJ6zfHjx7F9+3acP38e48ePx8qVK3H16lXo9XqMGzcOABATE4N33nnHYoI2O8Sxbds2FBcXw9/fHwqFAuPGjcOwYcOQmZmJoqIi+z4pEZEDdUCwetPpdKirq+uy6XS6TjEbGhrg7e1t2ler1aivrzftNzc3Y9SoUVi5ciX27t0LnU6Hd999t8v7vL29O73vdsz2oMvKyqDVaqFQKBAbG4vk5GTs3LkTsbGxePrpp/H0009b/ZdFRORMtgxwFBQUID8/v8vxlJQUpKam/hZT6BpVofjtmfL+/ftj+/btpv2kpCSsWrUK06ZNM/u+2zGboG/cuIHW1lZ4enpCr9fj559/BgB4enqin8hPBRIRiandhhSdkJCA6OjoLse9vLw67Ws0Ghw+fNi039DQALVabdq/ePEiqqqqEBcXB+BmQnd1dYVGo8HVq1dNr7ty5Uqn992O2SwbExODBQsW4O2330Z8fDxiYmJw4cIFxMbGIiIiwmJwIqLeItjwx8vLC8OHD++y3ZqgJ02ahOrqaly/fh2tra3Yv38/pk6dajrv7u6Ot99+G+fPn4cgCCgsLERISAh8fHzg5uaGI0eOAAC0Wm2n992O2R70Cy+8AD8/P5w4cQIZGRkIDg5Gc3Mz3nrrLdaFJiJJc8RNQo1Gg/T0dMTHx8NgMCAuLg7+/v5ITk5GWloa/Pz8kJOTgxdffBEGgwGPPvooFi9eDADYsGEDMjMz0dzcjEceeQTx8fEWr6cQuhtUcRBD/WlxAyrEH2ZxRLEkhXt/cQM6oliS0k3UcI4oGuQId2KxJIWrStR4ACC0t4keUzVstF3vX+xr/bJ8H9UW23UtR+E8aCKSJT7qTUQkUUbnDQ44DBM0EckSy43ayOPeGaLGazlTJmo8APC8f7boMVvOlosbsK1V3HgAPO+bKWq85uO7RY0HAAqVu+UX2UjQN4saT/Ti+gA8fUNFj9kXtLddsOv9jnjU29nYgyYiWeIYNBGRRHGIg4hIojjEQUQkUZzFQUQkUXfEEMeBAwdQXl6Oy5cvo1+/flCr1Zg6dSrCwsKc0T4ioh6R/U3CzZs349ixY5g7d66p8lJDQwM+++wzfPfdd1i5cqVTGklEZCvZj0GXlZXhv//7v7uUFo2IiEBERAQTNBFJlhyGOMxWG3Jzc8Ply5e7HL948SJUKvELrhARiUUQBKs3qTLbg87IyMCiRYvg6+trWq7lypUrqK2txdq1a53SQCKinjDKoAdtNkFPmjQJ5eXlOHbsGBoaGiAIAjQaDcaOHcseNBFJmhyGOMwm6IsXLwIAfHx84OPjYzr+69Itw4YNc2DTiIh6TspDF9Yym6CXLl2K2tpaqNXqLh9WoVDgiy++cGjjiIh6SvY96L///e9YuHAhsrKyEBgY6Kw2ERHZTQ7T7MzO4hgwYADWrFkDrVbrpOYQEYnDKAhWb1Jl8UlCf39/+Pv7O6MtRESikf0Qh+S59I3mK5TizngRBOk/xOqIhUnhohQ9pELsn6E+8jN5J3BUgi4tLcV7770Hg8GAxMRELFq0qNP5iooKbNmyBYIgYPjw4Vi7di0GDhwIrVaLDRs2YMiQIQCAJ554Aunp6WavxZ8mIpIlR8ziqK+vR15eHvbs2QOVSoX58+djwoQJGDlyJACgqakJ2dnZKC4uhkajwebNm7FlyxZkZmaipqYGGRkZiIiIsPp6ZsegiYj6qg4IVm/WqqqqwsSJEzFo0CB4enoiLCwM5eW/LWlnMBiQnZ0NjUYDAHjooYdw6dIlAEBNTQ20Wi3mzp2LV199Fb/88ovF6zFBE5EsCTb80el0qKur67LpdLpOMRsaGkxPVQOAWq1GfX29af+uu+7CzJk31/fU6/X44IMPTPve3t5ITU1FSUkJhg4dipycHIufgUMcRCRLRhvu1RQUFCA/P7/L8ZSUFKSmppr2uxs2USgUXY41NjbipZdewsMPP4zo6GgAwNatW03nlyxZYkrc5jBBE5Es2TIGnZCQYEqkv+fl5dVpX6PR4PDhw6b9hoYGUynm3x97/vnnMXHiRKxatQrAzYRdXFyMxMREU9tcXS2nXyZoIpIlW8aWvby8uiTj7kyaNAlbtmzB9evX4eHhgf379+PPf/6z6bzRaMSyZcsQHh6Ol156yXTc09MTH374IQICAjB27Fjs2rULISEhFq9nNkFbekDlqaeesngBIqLe4IgnCTUaDdLT0xEfHw+DwYC4uDj4+/sjOTkZaWlpuHz5Mk6cOAGj0Yh9+/YBAMaMGYPc3Fxs2rQJ2dnZ0Ov18PX1xfr16y1eTyGY+X9ARkYG9u3bh1mzZnV73taSo64qH8svskFL7X5R4wGAp2+o6DFbz1eKGk9o04saDwA8758taryWU3tFjQcAULqJHlL0+doOmAftMfwJ0WP2Be1tF+x6/xjNRKtfe7z+a7uu5Shmf5rWrVuHn3/+GYGBgYiLi3NWm4iI7Cb7WhwAkJOTY9V8PSIiKTEKHVZvUmVVPejw8HDT17/HetBEJFUdEi6CZC3WgyYiWZLDEAfrQRORLMmhB8160EQkS7Y86i1VrAdNRLJkFIy93QS78UlCIpIl2S8aK3nG9t5ugVUEQ5u4AQ03xI3nAEK7yJ8ZgKKf+MUXBYXIMSU8ZetOwxVViIgkij1oIiKJksMsDiZoIpIlKc/OsBYTNBHJkpQf4baW2Tsk7e3tKCgowLp16zoVqQaALVu2OLRhRET2EATB6k2qzCbo119/HSdPnoRarcYf//hHvP/++6ZzlZXiltAkIhJThyBYvUmV2SGO48eP47/+678A3CzOn5iYCHd3dyQmJkr6Xx0iIjnkKLMJWhAEtLS0wNPTE4MHD8b27duxYMECDBkypNuFEomIpEIO86DNDnE8++yziI6ORnV1NYCby71s374dGzduxJkzZ5zSQCKinpDDGLTZHvQzzzyDCRMmQKX6bVmg+++/H6Wlpfjss88c3jgiop6SwywOiwX7f03OtxbsDw0Vf+0+IiKxSPnmn7VYsJ+IZEnKQxfWYsF+IpIlRz1JWFpaivfeew8GgwGJiYlYtGhRp/MnT55EZmYmmpqaEBQUhDfeeAOurq64ePEiVqxYgWvXruEPf/gDNmzYgP79+5u9Fgv2E5EsOeImYX19PfLy8vC3v/0NJSUl2L17N/75z392es2KFSuwevVq7Nu3D4IgoKioCADwxhtvYOHChSgvL8eYMWPw7rvvWryexVqL/v7++POf/2z1ByAikgJbHlTR6XSoq6vrsul0uk4xq6qqMHHiRAwaNAienp4ICwtDeXm56fyFCxeg1+sxbtw4AEBMTAzKy8thMBjw7bffIiwsrNNxS5xai6O97YIzL9cjfaGNjnCnfu6+gN+bnrHl723Lli3Iz8/vcjwlJQWpqamm/YaGBnh7e5v21Wo1jh07dtvz3t7eqK+vx08//YQBAwbA1dW103FLWCyJiO54CQkJiI6O7nLcy8ur0353wyG/f2jvductve92mKCJ6I7n5eXVJRl3R6PRdCoc19DQALVa3en81atXTftXrlyBWq3G4MGD0dTUBKPRCBcXF9NxS8RfQ4iISKYmTZqE6upqXL9+Ha2trdi/fz+mTp1qOu/j4wM3NzccOXIEAKDVajF16lQolUoEBQWhrKys03FLFIIcJgsSETlJaWkptm3bBoPBgLi4OCQnJyM5ORlpaWnw8/PDqVOnkJmZiebmZjzyyCNYu3YtVCoVLly4gIyMDFy7dg1Dhw7Fxo0bMXDgQLPXYoImIpIoDnEQEUkUEzQRkUQxQRMRSRQTNBGRREkqQZeWlmL27NkICQlBYWGhaHGbmpoQERGBuro6u2Pl5+djzpw5mDNnDtavXy9C64DNmzdj9uzZmDNnDj766CNRYgLAW2+9hYyMDFFixcfHY86cOYiKikJUVBSOHj1qd8zKykrExMRg1qxZWLNmjd3xPv30U1P7oqKiEBgYiJycHLtilpSUmL7fb731lt1tBIAPPvgAYWFhiIyMxHvvvdfjOLf+XFdVVSEyMhKhoaHIy8sTJSYArFy5Env27BEl3u7duxEREYHIyEj86U9/QltbW4/i3jEEibh8+bIwffp04aeffhKam5uFyMhI4YcffrA77nfffSdEREQIo0ePFs6fP29XrEOHDgnPPPOMcOPGDaGtrU2Ij48X9u/fb1fM//3f/xXmz58vGAwGobW1VZg+fbpw5swZu2IKgiBUVVUJEyZMEFauXGl3rI6ODuHxxx8XDAaD3bF+9eOPPwqTJ08WLl26JLS1tQkLFiwQvvzyS9Hi/9///Z8QEhIiXLt2rccxWlpahPHjxwvXrl0TDAaDEBcXJxw6dMiudh06dEiIiIgQGhsbhfb2dmHp0qXCvn37bI5z6891a2urMG3aNOHHH38UDAaDkJSUZPPf560xL1++LCxdulTw9/cXiouL7W7j2bNnhZCQEKGxsVHo6OgQ/vjHPwofffSRzXHvJJLpQVsqQtJTRUVFyMrKsuqpHUu8vb2RkZEBlUoFpVKJ+++/v8tCBrZ67LHHsHPnTri6uuLatWswGo3w9PS0K+bPP/+MvLw8LFu2zK44vzp79iwUCgWSk5Mxd+5c7Nq1y+6Y//jHPzB79mzcc889UCqVyMvLw9ixY0Vo7U3Z2dlIT0/H4MGDexzDaDSio6MDra2taG9vR3t7O9zc3Oxq14kTJzB58mQMGDAALi4umDJlCioqKmyOc+vP9bFjx3Dffffh3nvvhaurKyIjI23+/bk1ZmlpKZ588kmEh4fb3L7u4qlUKmRnZ2PAgAFQKBR48MEH7f79kTvJPOptqQhJT+Xm5tod41cPPPCA6eva2lqUlZXhk08+sTuuUqnEO++8gx07dmDWrFnQaDR2xXv99deRnp6OS5cu2d02ANDpdAgODkZ2djb0ej3i4+Pxhz/8AY8//niPY547dw5KpRLPP/88rly5gunTp+OVV14Rpb1VVVXQ6/U9Tiy/GjBgAJYvX47w8HC4u7vjsccew6OPPmpXzNGjR+PNN9/E0qVL4eHhgcrKyh4Vlr/157q73x9rivGYi7lkyRIAMD0VZ28bfXx84OPjAwC4fv06CgsLsXbt2h7FvlNIpgfd3Q+pVFcO/+GHH5CUlISVK1fC19dXlJhpaWmorq7GpUuXTPVje+LTTz/F0KFDERwcLEq7ACAgIADr1683re4eFxeHr776yq6YRqMR1dXVePvtt1FUVISamhrs3btXlPZ+8sknWLx4sd1xTp06heLiYvzP//wPDh48iH79+uGvf/2rXTGDg4MRExOD5557DkuWLEFgYCCUSqXdbe1Lvz/19fVISEhAbGwsJkyY0NvNkTTJJOhbi4zcWoREKo4cOYLExET8x3/8R7fVr2x15swZnDx5EgDg4eGB0NBQnD59usfxysrKcOjQIURFReGdd95BZWUl3nzzTbvaePjwYdPK7sDNZPBr2cSeuvvuuxEcHIzBgwfD3d0dTz75pCj/Y2pra8O3336LGTNm2B3r4MGDCA4OxpAhQ6BSqRATE4NvvvnGrphNTU0ICQlBaWkpPv74Y3h4eODee++1u6195ffnzJkzWLBgAaKjo/Hyyy/3dnMkTzIJ2lIREim4dOkSXn75ZWzYsAFz5swRJWZdXR0yMzPR1taGtrY2fPHFF3YtL/bRRx/h888/R0lJCdLS0jBjxgysWrXKrjY2NjZi/fr1uHHjBpqamrB3716EhITYFXP69Ok4ePAgdDodjEYjDhw4gNGjR9sVEwBOnz4NX19fu8fxAeDhhx9GVVUVWlpaIAgCKisr4efnZ1fMuro6vPzyy2hvb0djYyM+/fRTu4diAGDs2LH417/+hXPnzsFoNOLzzz+X3O9PU1MTnn/+eSxfvhxJSUm93Zw+QTJj0BqNBunp6YiPjzcVIfH39+/tZnXy17/+FTdu3MC6detMx+bPn48FCxb0OOa0adNw9OhRPPXUU3BxcUFoaKhoyV8s06dPN7Wxo6MDCxcuREBAgF0xx44diyVLlmDhwoUwGAx4/PHHERsba3dbz58/j3vuucfuOAAwefJknDhxAjExMVAqlfDz88MLL7xgV8yHH34YoaGhmDt3LoxGIxITE0VZ79PNzQ3r1q1Damoqbty4gWnTpmHWrFl2xxXTZ599hqtXr2LHjh3YsWMHAGDGjBlYvnx5L7dMulgsiYhIoiQzxEFERJ0xQRMRSRQTNBGRRDFBExFJFBM0EZFEMUETEUkUEzQRkUQxQRMRSdT/B6uRph1918kPAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_doc_similarity(lda_fit, doc_term_matrix)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Model Performance\n",
    "To assess the LDA models performance, the perplexity and coherence can be computed for a hold out test corpus. These measures can also be used to determine the number of topics to use in the model. Perplexity (actually log perplexity) is an entropy-like measure based on the likelihood of a topic model. That is, the lower the perplexity, the better a model fits the train or test data. However, a model with a low perplexity is not necessarily the one that humans can interpret best. Because of that, the coherence of words within a topic can be a more meaningful measure to select the number of topics. Coherence can be measured in different ways (see Röder, Both, and Hinneburg, 2015), but most of them are based on co-occurence (how often do words belonging to the same topic occur in the same document, sentence, etc.). The measure $C_{UMass}$ used here is also on the log-scale so it is negative. In contrast to perplexity, higher coherence means a model with topics that are easier to interpret."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to evaluate the LDA model performance using hold out perplexity and coherence\n",
    "\n",
    "def calc_LDA_performance(corpus, K, train_test_split = 0.8, seed = None):\n",
    "    if seed is not None:\n",
    "        random.seed(seed)\n",
    "    \n",
    "    dictionary = corpora.Dictionary(corpus)\n",
    "    doc_term_matrix = [dictionary.doc2bow(doc) for doc in corpus]\n",
    "    \n",
    "    doc_idx = range(len(corpus))\n",
    "    train_idx = random.sample(doc_idx, k = math.floor(train_test_split*(len(corpus))))\n",
    "    test_idx = [doc_idx[i] for i in doc_idx if i not in train_idx]\n",
    "    train_doc_term_matrix = [doc_term_matrix[i] for i in train_idx]\n",
    "    test_doc_term_matrix = [doc_term_matrix[i] for i in test_idx] if type(test_idx) is list else [doc_term_matrix[test_idx]]\n",
    "    \n",
    "    performance = {\"k\": [], \"train_perplexity\": [], \"test_perplexity\": [], \"train_coherence\": [], \"test_coherence\": []}\n",
    "\n",
    "    for k in K: \n",
    "        train_lda_fit = LdaModel(train_doc_term_matrix, num_topics = k, id2word = dictionary, passes = 10, random_state = seed)\n",
    "        train_perplexity = train_lda_fit.log_perplexity(train_doc_term_matrix)\n",
    "        test_perplexity = train_lda_fit.log_perplexity(test_doc_term_matrix)\n",
    "        \n",
    "        performance[\"k\"].append(k)\n",
    "        performance[\"train_perplexity\"].append(train_perplexity)\n",
    "        performance[\"test_perplexity\"].append(test_perplexity)\n",
    "        \n",
    "        train_coherence_model_fit = CoherenceModel(\n",
    "            model = train_lda_fit,\n",
    "            texts = [corpus[i] for i in train_idx],\n",
    "            corpus = train_doc_term_matrix,\n",
    "            dictionary = dictionary,\n",
    "            coherence = \"c_v\"\n",
    "        )\n",
    "        test_coherence_model_fit = CoherenceModel(\n",
    "            model = train_lda_fit,\n",
    "            texts = [corpus[i] for i in test_idx] if type(test_idx) is list else corpus[test_idx],\n",
    "            corpus = test_doc_term_matrix,\n",
    "            dictionary = dictionary,\n",
    "            coherence = \"u_mass\"\n",
    "        )\n",
    "        \n",
    "        performance[\"train_coherence\"].append(train_coherence_model_fit.get_coherence())\n",
    "        performance[\"test_coherence\"].append(test_coherence_model_fit.get_coherence())\n",
    "    \n",
    "    return pd.DataFrame(performance)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "    k  train_perplexity  test_perplexity  train_coherence  test_coherence\n",
      "0   2         -7.186948        -9.598770         0.281702       -8.799914\n",
      "1   3         -7.174706       -10.809252         0.284994       -6.989347\n",
      "2   4         -7.236096       -11.715559         0.328435       -9.369097\n",
      "3   5         -7.254988       -12.553841         0.316915       -8.690004\n",
      "4   6         -7.299072       -13.365428         0.400858       -7.414696\n",
      "5   7         -7.342455       -13.872166         0.378040       -8.549767\n",
      "6   8         -7.318261       -14.803246         0.419115       -8.185515\n",
      "7   9         -7.370012       -15.538190         0.390334       -7.930280\n",
      "8  10         -7.287715       -16.720492         0.494777       -9.353981\n"
     ]
    }
   ],
   "source": [
    "# Evaluate model performance\n",
    "\n",
    "K = range(2, 11)\n",
    "\n",
    "lda_performance_df = calc_LDA_performance(corpus, K, seed = SEED)\n",
    "\n",
    "print(lda_performance_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to plot model performance\n",
    "\n",
    "def plot_lda_performance(performance_df):\n",
    "    plot_df = performance_df.melt(id_vars = \"k\", var_name = \"measure\")\n",
    "    plot_df[[\"set\", \"measure\"]] = [s.split(\"_\") for s in plot_df.measure]\n",
    "    \n",
    "    plot = sns.relplot(\n",
    "        data = plot_df,\n",
    "        x = \"k\",\n",
    "        y = \"value\",\n",
    "        row = \"measure\",\n",
    "        style = \"set\",\n",
    "        kind = \"line\",\n",
    "        facet_kws = {'sharey': False, 'sharex': True}\n",
    "    )\n",
    "    \n",
    "    return plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<seaborn.axisgrid.FacetGrid at 0x25839c064c0>"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 419.375x720 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot model performance\n",
    "\n",
    "plot_lda_performance(lda_performance_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Suggestions for Future Work\n",
    "\n",
    "1. Look at tenses (past vs. future)\n",
    "2. Look at sentiment"
   ]
  }
 ],
 "metadata": {
  "interpreter": {
   "hash": "7347d991c5e2f391a3d7d79e5ee26bb4d126111ca53b3542519de53e7706b949"
  },
  "kernelspec": {
   "display_name": "navigating-stories",
   "language": "python",
   "name": "navigating-stories"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}