LDA.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "98c5c006",
   "metadata": {},
   "source": [
    "# Latent Dirichlet Allocation (LDA) Modelling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "bd0a91c8",
   "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>date</th>\n",
       "      <th>like_count</th>\n",
       "      <th>reply_count</th>\n",
       "      <th>retweet_count</th>\n",
       "      <th>text</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2011-12-01</td>\n",
       "      <td>267</td>\n",
       "      <td>63</td>\n",
       "      <td>24</td>\n",
       "      <td>{I made the volume on the Model S  http://t.co...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2011-12-03</td>\n",
       "      <td>82</td>\n",
       "      <td>38</td>\n",
       "      <td>31</td>\n",
       "      <td>{That was a total non sequitur btw, Great Volt...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2011-12-04</td>\n",
       "      <td>65</td>\n",
       "      <td>17</td>\n",
       "      <td>9</td>\n",
       "      <td>{Am reading a great biography of Ben Franklin ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2011-12-21</td>\n",
       "      <td>1330</td>\n",
       "      <td>87</td>\n",
       "      <td>597</td>\n",
       "      <td>{Yum! Even better than deep fried butter:  htt...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2011-12-22</td>\n",
       "      <td>1349</td>\n",
       "      <td>132</td>\n",
       "      <td>206</td>\n",
       "      <td>{Model S options are out! Performance in red a...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         date  like_count  reply_count  retweet_count  \\\n",
       "0  2011-12-01         267           63             24   \n",
       "1  2011-12-03          82           38             31   \n",
       "2  2011-12-04          65           17              9   \n",
       "3  2011-12-21        1330           87            597   \n",
       "4  2011-12-22        1349          132            206   \n",
       "\n",
       "                                                text  \n",
       "0  {I made the volume on the Model S  http://t.co...  \n",
       "1  {That was a total non sequitur btw, Great Volt...  \n",
       "2  {Am reading a great biography of Ben Franklin ...  \n",
       "3  {Yum! Even better than deep fried butter:  htt...  \n",
       "4  {Model S options are out! Performance in red a...  "
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Load to the data\n",
    "import pandas as pd\n",
    "df_tweets = pd.read_csv('data/tweets_data_lda.csv')\n",
    "df_tweets.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "dffae5f8",
   "metadata": {},
   "source": [
    "## Data Pre-processing\n",
    "**Preprocess the data by making it all lowercase. Remove a reasonable set of stopwords from the dataset and tokenize. Then, report the 10 most common words and their count. We need to iterate this process, adding some stop words as we understand the structure of the data. Justify additional stop words we've added.**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "441873e6",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[nltk_data] Downloading package stopwords to /Users/ziza/nltk_data...\n",
      "[nltk_data]   Package stopwords is already up-to-date!\n"
     ]
    },
    {
     "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>date</th>\n",
       "      <th>like_count</th>\n",
       "      <th>reply_count</th>\n",
       "      <th>retweet_count</th>\n",
       "      <th>text</th>\n",
       "      <th>preprocessed_text</th>\n",
       "      <th>tokenized_text</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2011-12-01</td>\n",
       "      <td>267</td>\n",
       "      <td>63</td>\n",
       "      <td>24</td>\n",
       "      <td>{I made the volume on the Model S  http://t.co...</td>\n",
       "      <td>{i made the volume on the model s go to 11. no...</td>\n",
       "      <td>[made, volume, model, go, need, work, miniatur...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2011-12-03</td>\n",
       "      <td>82</td>\n",
       "      <td>38</td>\n",
       "      <td>31</td>\n",
       "      <td>{That was a total non sequitur btw, Great Volt...</td>\n",
       "      <td>{that was a total non sequitur btw, great volt...</td>\n",
       "      <td>[total, non, sequitur, great, voltaire, quote,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2011-12-04</td>\n",
       "      <td>65</td>\n",
       "      <td>17</td>\n",
       "      <td>9</td>\n",
       "      <td>{Am reading a great biography of Ben Franklin ...</td>\n",
       "      <td>{am reading a great biography of ben franklin ...</td>\n",
       "      <td>[reading, great, biography, ben, franklin, isa...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2011-12-21</td>\n",
       "      <td>1330</td>\n",
       "      <td>87</td>\n",
       "      <td>597</td>\n",
       "      <td>{Yum! Even better than deep fried butter:  htt...</td>\n",
       "      <td>{yum! even better than deep fried butter: yeah...</td>\n",
       "      <td>[yum, even, better, deep, fried, butter, yeah,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2011-12-22</td>\n",
       "      <td>1349</td>\n",
       "      <td>132</td>\n",
       "      <td>206</td>\n",
       "      <td>{Model S options are out! Performance in red a...</td>\n",
       "      <td>{model s options are out! performance in red a...</td>\n",
       "      <td>[model, options, performance, red, black, deli...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         date  like_count  reply_count  retweet_count  \\\n",
       "0  2011-12-01         267           63             24   \n",
       "1  2011-12-03          82           38             31   \n",
       "2  2011-12-04          65           17              9   \n",
       "3  2011-12-21        1330           87            597   \n",
       "4  2011-12-22        1349          132            206   \n",
       "\n",
       "                                                text  \\\n",
       "0  {I made the volume on the Model S  http://t.co...   \n",
       "1  {That was a total non sequitur btw, Great Volt...   \n",
       "2  {Am reading a great biography of Ben Franklin ...   \n",
       "3  {Yum! Even better than deep fried butter:  htt...   \n",
       "4  {Model S options are out! Performance in red a...   \n",
       "\n",
       "                                   preprocessed_text  \\\n",
       "0  {i made the volume on the model s go to 11. no...   \n",
       "1  {that was a total non sequitur btw, great volt...   \n",
       "2  {am reading a great biography of ben franklin ...   \n",
       "3  {yum! even better than deep fried butter: yeah...   \n",
       "4  {model s options are out! performance in red a...   \n",
       "\n",
       "                                      tokenized_text  \n",
       "0  [made, volume, model, go, need, work, miniatur...  \n",
       "1  [total, non, sequitur, great, voltaire, quote,...  \n",
       "2  [reading, great, biography, ben, franklin, isa...  \n",
       "3  [yum, even, better, deep, fried, butter, yeah,...  \n",
       "4  [model, options, performance, red, black, deli...  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Data Pre-processing and make the tweets all lowercase and remove stopwords.\n",
    "from sklearn.feature_extraction.text import CountVectorizer\n",
    "from gensim.corpora import Dictionary\n",
    "from gensim.models.ldamodel import LdaModel\n",
    "from gensim.models import CoherenceModel\n",
    "from nltk.corpus import stopwords\n",
    "from nltk.tokenize import RegexpTokenizer\n",
    "from datetime import datetime\n",
    "from nltk.stem import PorterStemmer\n",
    "import re\n",
    "import nltk\n",
    "nltk.download('stopwords')\n",
    "import math\n",
    "import math\n",
    "\n",
    "twitter_df = df_tweets.copy()\n",
    "\n",
    "# lower the tweets\n",
    "twitter_df['preprocessed_text'] = twitter_df['text'].str.lower()\n",
    "\n",
    "# remove apostrophe from words and url\n",
    "twitter_df['preprocessed_text'] = [re.sub(\"('[a-z]+)\\s\", \" \", row) for row in twitter_df['preprocessed_text']]\n",
    "twitter_df['preprocessed_text'] = [re.sub(\"(')\\s\", \" \", row) for row in twitter_df['preprocessed_text']]\n",
    "twitter_df['preprocessed_text'] = [re.sub(\"(?:https:\\/\\/\\S+)\\s\", \"\", row) for row in twitter_df['preprocessed_text']]\n",
    "\n",
    "                                      \n",
    "# filter out rest URLs\n",
    "url_re = '(?:https?:\\/\\/)?(?:[^?\\/\\s]+[?\\/])(.*)'\n",
    "twitter_df['preprocessed_text'] = twitter_df['preprocessed_text'].apply(lambda row: ' '.join([word for word in row.split() if (not re.match(url_re, word))]))\n",
    "\n",
    "# tokenize the tweets\n",
    "tokenizer = RegexpTokenizer('[a-zA-Z]\\w+\\'?\\w*')\n",
    "twitter_df['tokenized_text'] = twitter_df['preprocessed_text'].apply(lambda row: tokenizer.tokenize(row))\n",
    "\n",
    "#create an object of class PorterStemmer\n",
    "porter = PorterStemmer()\n",
    "\n",
    "# apply stemming\n",
    "twitter_df['preprocessed_text'] = [porter.stem(row) for row in twitter_df['preprocessed_text']]   \n",
    "\n",
    "# filter out stop words\n",
    "en_stop_words = nltk.corpus.stopwords.words('english')\n",
    "additional_stop_words =['amp', 'rt', 'th','co', 're', 've', 'kim', 'daca', 'us', 'it', 'th', 'you', 'haha', 'st', 'et', 'so', 'iii', 'also', 've', 'la', 're', 'the', 'https', 'wow', 'actually', 'due', 'ft', 'pcr', 'via', 'am', 'gt', 'com', 'since', 'in', 'me', 'and', 'btw', 'yesterday', 'ii', 'inu', 'on', 'http', 'to', 'vs', 'rd', 'ur', 'of', 'bs', 'km', 'est', 'em', 'lz', 'kms', 'aft', 'nd',  'here’s', 're', 'mqxfakpzf' 'mph', 'ht', 'etc', 'dm', 'doo']\n",
    "en_stop_words.extend(additional_stop_words)\n",
    "\n",
    "twitter_df['tokenized_text'] = twitter_df['tokenized_text'].apply(lambda row: [word for word in row if (not word in en_stop_words)])\n",
    "\n",
    "df_tweets_clean = twitter_df.copy()\n",
    "df_tweets_clean.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8a2d3c3a",
   "metadata": {},
   "source": [
    "### Bag of Words\n",
    "If we want to classify text on a topic basis, we often do not want to look at the sequences of words. We do not want to classify text. Instead, we present the text as an unordered word package, ignoring its original position in the text and keeping its frequency."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "dd9e3757",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# define function for counting words\n",
    "def get_most_freq_words(str, n=None):\n",
    "    vect = CountVectorizer().fit(str)\n",
    "    bag_of_words = vect.transform(str)\n",
    "    sum_words = bag_of_words.sum(axis=0) \n",
    "    freq = [(word, sum_words[0, idx]) for word, idx in vect.vocabulary_.items()]\n",
    "    freq =sorted(freq, key = lambda x: x[1], reverse=True)\n",
    "    return freq[:n]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "0d2520cc",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Elon Musked used 8197 words in his tweets.\n",
      "The most popular words:\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[('tesla', 629),\n",
       " ('model', 260),\n",
       " ('spacex', 245),\n",
       " ('good', 180),\n",
       " ('launch', 174),\n",
       " ('rocket', 166),\n",
       " ('car', 160),\n",
       " ('falcon', 159),\n",
       " ('like', 151),\n",
       " ('time', 129)]"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# count words in tweets\n",
    "words_in_tweets = get_most_freq_words([ word for tweet in df_tweets_clean.tokenized_text for word in tweet])\n",
    "print(f'Elon Musked used {len(words_in_tweets)} words in his tweets.')\n",
    "print(f'The most popular words:')\n",
    "words_in_tweets[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "22f7208d",
   "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>Word</th>\n",
       "      <th>Count</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>tesla</td>\n",
       "      <td>629</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>model</td>\n",
       "      <td>260</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>spacex</td>\n",
       "      <td>245</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>good</td>\n",
       "      <td>180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>launch</td>\n",
       "      <td>174</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "     Word  Count\n",
       "0   tesla    629\n",
       "1   model    260\n",
       "2  spacex    245\n",
       "3    good    180\n",
       "4  launch    174"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Convert to DataFrame\n",
    "words_count = pd.DataFrame(words_in_tweets, columns=['Word', 'Count'])\n",
    "words_count.head()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "14e06c7e",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Export word counts to csv\n",
    "words_count.to_csv('data/words_count.csv', index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a45294f4",
   "metadata": {},
   "source": [
    "### Identify the number of subjects"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "5440be22",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1 -2.486770372323403\n",
      "2 -2.4624495886762334\n",
      "3 -2.6362883481127546\n",
      "4 -2.75211167009597\n",
      "5 -3.1964512647143315\n",
      "6 -2.8829077168705397\n",
      "7 -2.8939529398438046\n",
      "8 -3.058546237527238\n",
      "9 -3.547223286936251\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 720x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# build a dictionary where for each tweet, each word has its own id.\n",
    "tweets_dictionary = Dictionary(df_tweets_clean.tokenized_text)\n",
    "\n",
    "# build the corpus i.e. vectors with the number of occurence of each word per tweet\n",
    "tweets_corpus = [tweets_dictionary.doc2bow(tweet) for tweet in df_tweets_clean.tokenized_text]\n",
    "\n",
    "# compute coherence\n",
    "tweets_coherence = []\n",
    "for nb_topics in range(1,10):\n",
    "    lda = LdaModel(tweets_corpus, num_topics = nb_topics, alpha= 'auto', eta='auto', id2word = tweets_dictionary, passes=10)\n",
    "    cohm = CoherenceModel(model=lda, corpus=tweets_corpus, dictionary=tweets_dictionary, coherence='u_mass')\n",
    "    coh = cohm.get_coherence()\n",
    "    tweets_coherence.append(coh)\n",
    "    print(nb_topics,coh)\n",
    "\n",
    "# visualize coherence\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "plt.figure(figsize=(10,5))\n",
    "plt.plot(range(1, 10),tweets_coherence)\n",
    "plt.xlabel(\"Number of Topics\")\n",
    "plt.ylabel(\"Coherence Score\");"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "654d40ed",
   "metadata": {},
   "source": [
    "The average coherence score per topic for a variety of models trained with varying numbers of topics is shown below. The number of topics for which the average score reaches a peak point is the perfect balance we seek. As a result, our best guess for the number of topics is around two (k =2)."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "50820618",
   "metadata": {},
   "source": [
    "### Running the LDA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "ba231ad7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1440x1080 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.gridspec as gridspec\n",
    "\n",
    "k = 2\n",
    "tweets_lda = LdaModel(tweets_corpus, num_topics = k, id2word = tweets_dictionary, passes=10)\n",
    "\n",
    "def plot_top_words(lda=tweets_lda, nb_topics=k, nb_words=7):\n",
    "    top_words = [[word for word,_ in lda.show_topic(topic_id, topn=50)] for topic_id in range(lda.num_topics)]\n",
    "    top_betas = [[beta for _,beta in lda.show_topic(topic_id, topn=50)] for topic_id in range(lda.num_topics)]\n",
    "\n",
    "    gs  = gridspec.GridSpec(round(math.sqrt(k))+1,round(math.sqrt(k))+1)\n",
    "    gs.update(wspace=0.5, hspace=0.5)\n",
    "    plt.figure(figsize=(20,15))\n",
    "    for i in range(nb_topics):\n",
    "        ax = plt.subplot(gs[i])\n",
    "        plt.barh(range(nb_words), top_betas[i][:nb_words], align='center',color='pink', ecolor='black')\n",
    "        ax.invert_yaxis()\n",
    "        ax.set_yticks(range(nb_words))\n",
    "        ax.set_yticklabels(top_words[i][:nb_words])\n",
    "        plt.title(\"Topic \"+str(i))\n",
    "        \n",
    "  \n",
    "plot_top_words()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a53f203e",
   "metadata": {},
   "source": [
    "### We can determine topics based on key words :\n",
    "### Topic 0 - SpaceX\n",
    "### Topic 1 - Tesla"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "338d658e",
   "metadata": {},
   "source": [
    "Reference; https://towardsdatascience.com/topic-modeling-with-latent-dirichlet-allocation-by-example-3b22cd10c835, date; 2021-07-21, time; 2.25 pm"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "mlenv",
   "language": "python",
   "name": "mlenv"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}