topic_model_example.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T22:31:58.474241Z",
     "start_time": "2018-12-17T22:31:56.298343Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style>.container {width:100% !important;} </style>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\")\n",
    "\n",
    "import pandas as pd \n",
    "import numpy as np\n",
    "import re\n",
    "import string\n",
    "import collections\n",
    "import matplotlib.pyplot as plt\n",
    "import time \n",
    "\n",
    "import sys\n",
    "sys.path.insert(0, 'helpers/')\n",
    "from helper_base import *\n",
    "from helper_prep import *\n",
    "from helper_model import *\n",
    "\n",
    "from gensim.models import Word2Vec"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T16:20:03.940303Z",
     "start_time": "2018-12-13T16:20:03.099768Z"
    }
   },
   "outputs": [],
   "source": [
    "df = pd.read_csv('../df_output_20181213_1004.csv')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Below we create a lemmatized version of the text field since our n-grams were identified on lemmas:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-14T18:20:38.753975Z",
     "start_time": "2018-12-14T18:17:57.716842Z"
    }
   },
   "outputs": [],
   "source": [
    "df['tokenized_text'] = df.reviews_regram.apply(lambda x: ' '.join(retokenize(x))) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-14T18:20:38.787184Z",
     "start_time": "2018-12-14T18:20:38.756057Z"
    },
    "scrolled": true
   },
   "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>reviewerID</th>\n",
       "      <th>asin</th>\n",
       "      <th>reviewerName</th>\n",
       "      <th>helpful</th>\n",
       "      <th>reviewText</th>\n",
       "      <th>overall</th>\n",
       "      <th>summary</th>\n",
       "      <th>unixReviewTime</th>\n",
       "      <th>reviewTime</th>\n",
       "      <th>review_year</th>\n",
       "      <th>reviews_clean</th>\n",
       "      <th>reviews_regram</th>\n",
       "      <th>tokenized_text</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>A1VEELTKS8NLZB</td>\n",
       "      <td>616719923X</td>\n",
       "      <td>Amazon Customer</td>\n",
       "      <td>[0, 0]</td>\n",
       "      <td>Just another flavor of Kit Kat but the taste i...</td>\n",
       "      <td>4.0</td>\n",
       "      <td>Good Taste</td>\n",
       "      <td>1370044800</td>\n",
       "      <td>06 1, 2013</td>\n",
       "      <td>2013</td>\n",
       "      <td>just another flavor of kit kat but the taste i...</td>\n",
       "      <td>just another flavor of kit kat but the taste i...</td>\n",
       "      <td>another flavor kit kat taste unique bit differ...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       reviewerID        asin     reviewerName helpful  \\\n",
       "0  A1VEELTKS8NLZB  616719923X  Amazon Customer  [0, 0]   \n",
       "\n",
       "                                          reviewText  overall     summary  \\\n",
       "0  Just another flavor of Kit Kat but the taste i...      4.0  Good Taste   \n",
       "\n",
       "   unixReviewTime  reviewTime  review_year  \\\n",
       "0      1370044800  06 1, 2013         2013   \n",
       "\n",
       "                                       reviews_clean  \\\n",
       "0  just another flavor of kit kat but the taste i...   \n",
       "\n",
       "                                      reviews_regram  \\\n",
       "0  just another flavor of kit kat but the taste i...   \n",
       "\n",
       "                                      tokenized_text  \n",
       "0  another flavor kit kat taste unique bit differ...  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "display_all(df[:1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T16:24:24.846460Z",
     "start_time": "2018-12-13T16:24:24.842501Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(51894, 12)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Vectorization"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The function `get_vecs()` applies both count vectorization and TF-IDF vectorization to the text field. Both variations can then be passed into the topic models for consideration. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T16:51:24.069094Z",
     "start_time": "2018-12-13T16:45:49.228805Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Num. Features (Count Vec):  46891\n",
      "Num. Features (TFIDF Vec):  46891\n",
      "\n",
      "Dim. Docs (Count Vec):  (51894, 46891)\n",
      "Dim. Docs (TFIDF Vec):  (51894, 46891)\n"
     ]
    }
   ],
   "source": [
    "sentences = df.reviews_regram\n",
    "cv_features, cv_docs, tf_features, tf_docs = get_vecs(sentences)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Below we try two topic models: Latent Dirichlet Allocation-based modelling using [gensim's LDA model](https://radimrehurek.com/gensim/models/ldamodel.html), as well as Greg Ver Steeg's [Correlation Explanation (CorEx)](https://github.com/gregversteeg/corex_topic) model. We found that CorEx was more effective than LDA on our survey data, possibly due to the small data size. It also provides a simple way to seed/anchor tokens of interest. \n",
    "\n",
    "The functions `get_gensim_topics()` and `get_corex_topics()` run both models using grid search on the hyperparameter `n`, the number of topics to generate, and calculate the key metric for each variation - coherence for LDA and correlation for CorEx. `pick_top_n()` then finds the hyperparameter(s) `n` that maximize(s) coherence/correlation so that we can view the optimal model(s) output.\n",
    "\n",
    "For grid search, we used `list(range(12,32,2))`, though this range is pretty variable. Just note that perplexity monotonically decreases as `n` increases, coherence goes up and down, and higher `n` is obviously harder to interpret by eye. For the purposes of this demo, we'll apply grid search on a smaller range. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## LDA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T17:13:31.275889Z",
     "start_time": "2018-12-13T17:06:27.218264Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "num_topics_gensim = list(range(12,22,2))\n",
    "perplexity, coherence = get_gensim_topics(num_topics_gensim, sentences)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T17:16:43.954044Z",
     "start_time": "2018-12-13T17:16:43.865659Z"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(num_topics_gensim, perplexity)\n",
    "plt.title('Perplexity Plot')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T17:16:55.728882Z",
     "start_time": "2018-12-13T17:16:55.644591Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(num_topics_gensim, coherence)\n",
    "plt.title('Coherence Plot')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T17:17:27.991969Z",
     "start_time": "2018-12-13T17:17:27.988547Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Num. Topics:  [12]\n"
     ]
    }
   ],
   "source": [
    "top_n = pick_top_n(coherence, num_topics_gensim, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T17:21:10.612744Z",
     "start_time": "2018-12-13T17:17:28.673018Z"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Num. Topics:  12\n",
      "\n",
      "cooky , make , sugar , add , cookie , use , water , cup , mix , sweet\n",
      "popcorn , salt , bag , package , product , use , open , pop , would , corn\n",
      "fat , ingredient , calorie , oil , use , coconut_oil , product , contain , organic , gram\n",
      "price , buy , product , find , use , love , great , make , amazon , good\n",
      "flavor , like , taste , try , would , one , good , really , drink , chocolate\n",
      "use , tuna , make , work , easy , great , well , mix , come , container\n",
      "snack , love , one , like , bag , get , eat , buy , good , box\n",
      "like , taste , eat , would , good , try , think , bar , really , flavor\n",
      "rice , add , soup , sauce , flavor , cook , make , garlic , chicken , couscous\n",
      "coffee , cup , brew , kcup , keurig , blend , filter , use , machine , k\n",
      "tea , like , taste , flavor , make , lipton , one , use , try , get\n",
      "get , take , day , like , go , use , try , put , milk , one\n",
      "\n"
     ]
    }
   ],
   "source": [
    "_,_ = get_gensim_topics(top_n, sentences, print_flag = True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T17:23:16.328281Z",
     "start_time": "2018-12-13T17:23:16.323732Z"
    }
   },
   "source": [
    "We can already see that we have more or less cohesive topics around: cookies, popcorn, oil, coffee, tea"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Corex with Count Vectorization"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Below, we run a version of the CorEx model using the count vectorization method. Note that you can also run a version with TF-IDF vectorization, using `tf_features` and `tf_docs` created above, and compare outputs. We generally found that as we were mostly interested in finding topics that were most representative in terms of volume/frequency, count vectorization was more effective.\n",
    "\n",
    "Let's use the same set of numbers we used in the gensim model for CorEx grid search."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T17:32:49.083556Z",
     "start_time": "2018-12-13T17:23:26.696884Z"
    }
   },
   "outputs": [],
   "source": [
    "correlation = get_corex_topics(num_topics_gensim, cv_docs, cv_features)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T17:33:12.854290Z",
     "start_time": "2018-12-13T17:33:12.765259Z"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(num_topics_gensim, correlation)\n",
    "plt.title('Correlation Plot')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T17:33:26.887587Z",
     "start_time": "2018-12-13T17:33:26.884037Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Num. Topics:  [20]\n"
     ]
    }
   ],
   "source": [
    "top_n_corex = pick_top_n(correlation, num_topics_gensim, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T17:35:31.365158Z",
     "start_time": "2018-12-13T17:33:29.742344Z"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Num topics:  20\n",
      "1: like, would, try, taste, think, one, say, first, much, bit\n",
      "2: coffee, tea, cup, brew, drink, strong, kcup, keurig, kcups, green_tea\n",
      "3: calorie, fat, ingredient, contain, gram, per, sugar, serving, protein, sodium\n",
      "4: oil, skin, coconut_oil, vitamin, help, calcium, iron, health, hair, process\n",
      "5: sauce, chicken, soup, cook, garlic, dish, cheese, onion, pepper, noodle\n",
      "6: add, minute, make, water, easy, microwave, boil, direction, prepare, let\n",
      "7: formula, pad, cleaner, cook_top, handle, enfamil, cleaning, baby, infant, scrub\n",
      "8: review, product, star, come, however, since, read, note, plastic, another\n",
      "9: snack, chocolate, eat, apple, bar, sweet, crunchy, cooky, fruit, nut\n",
      "10: open, bag, package, take, get, two, keep, container, small, size\n",
      "11: company, new, acid, system, become, show, oz, consumer, bean, american\n",
      "12: amazon, price, save, subscribe, buy, find, store, pay, local, sell\n",
      "13: use, mix, recipe, bake, butter, milk, powder, egg, popcorn, dry\n",
      "14: gluten, pasta, free, flour, barilla, wheat, texture, bob, red_mill, spaghetti\n",
      "15: year, time, day, every, ago, last, never, back, month, week\n",
      "16: honey, stevia, sweeten, sweetener, ginger, orange, lemon, juice, cold, sweetness\n",
      "17: go, box, three, old, kid, pack, child, tell, us, hi\n",
      "18: flavor, without, nice, color, prefer, enjoy, light, mild, variety, pleasant\n",
      "19: little, less, piece, amount, stick, hard, bite, regular, thick, hold\n",
      "20: also, recommend, many, feel, anything, lot, highly, type, energy, addition\n",
      "\n"
     ]
    }
   ],
   "source": [
    "_ = get_corex_topics(top_n_corex, cv_docs, cv_features, print_flag = True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can see that there are topics that emerge around coffee/tea, nutritional content, preparations, prices, pasta, and sweetener."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Anchored CorEx"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can use the words outputted from LDA and CorEx to anchor a CorEx model with the following specifications:\n",
    "- 8 topics are anchored using a maximum of 6 words per topic\n",
    "- `n_hidden=12` means the model will try to elicit 12 topics; this also means the model can find up to 4 topics unrelated to the anchors\n",
    "- `anchor_strength=4` anchors a topic relatively strongly to the words\n",
    "\n",
    "See the section Semi-Supervised Topic Modelling on [this](https://github.com/gregversteeg/corex_topic) page for more details."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T17:59:48.733254Z",
     "start_time": "2018-12-13T17:58:03.919437Z"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1: coffee, cup, drink, tea, brew, keurig, strong, kcup, kcups, green_tea\n",
      "2: ingredient, calorie, fat, serving, sodium, gram, contain, per, serve, protein\n",
      "3: add, minute, prepare, microwave, boil, direction, cook, dish, easy, chicken\n",
      "4: price, save, pay, subscribe, local, amazon, money, cost, cheaper, reasonably\n",
      "5: gluten, pasta, flour, wheat, barilla, spaghetti, free, sauce, bread, recipe\n",
      "6: honey, sweetener, sweeten, stevia, natural, help, effect, bottle, health, benefit\n",
      "7: sugar, mix, cooky, cookie, milk, keebler, chocolate_chip, shortbread, pancake, vanilla\n",
      "8: bag, package, salt, popcorn, pop, corn, kernel, salty, resealable, orville\n",
      "9: calorie, oil, fat, coconut_oil, saturate, trans, cholesterol, skin, vitamin, olive_oil\n",
      "10: one, get, go, first, even, could, know, take, thing, see\n",
      "11: flavor, note, nice, side, slightly, enjoy, type, mild, color, either\n",
      "12: would, try, think, taste, say, give, much, bit, something, review\n",
      "13: use, make, water, also, well, open, top, put, time, need\n",
      "14: snack, eat, apple, crunchy, chocolate, bar, fruit, sweet, nut, cracker\n",
      "15: like, product, look, food, really, less, seem, lot, include, label\n",
      "16: find, year, two, come, since, day, never, back, ago, brand\n"
     ]
    }
   ],
   "source": [
    "topic_model = ct.Corex(n_hidden=16, seed = 10)  \n",
    "topic_model.fit(cv_docs, words=cv_features,\\\n",
    "                anchors = [['coffee', 'tea', 'cup', 'brew', 'drink', 'keurig'],\\\n",
    "                           ['calorie', 'fat', 'ingredient', 'serving', 'sodium'],\\\n",
    "                           ['add', 'minute', 'microwave', 'boil', 'direction', 'prepare'],\\\n",
    "                           ['price', 'save', 'subscribe', 'pay', 'local'],\\\n",
    "                           ['gluten', 'pasta', 'flour', 'barilla', 'wheat', 'spaghetti'],\\\n",
    "                           ['honey', 'stevia', 'sweeten', 'sweetener'],\\\n",
    "                           ['cooky', 'cookie', 'sugar', 'mix'],\\\n",
    "                           ['popcorn', 'pop', 'corn', 'salt', 'bag', 'package'],\\\n",
    "                           ['fat', 'calorie', 'oil', 'coconut_oil']\\\n",
    "                          ], anchor_strength=4) #, docs=docs))\n",
    "\n",
    "topics = topic_model.get_topics()\n",
    "for topic_n,topic in enumerate(topics):\n",
    "    words,mis = zip(*topic)\n",
    "    topic_str = str(topic_n+1)+': '+', '.join(words)\n",
    "    print(topic_str)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Custom Word2Vec Model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-13T21:00:09.949294Z",
     "start_time": "2018-12-13T21:00:09.946102Z"
    }
   },
   "source": [
    "Let's say we're happy with the ten more-or-less coherent topics that emerge above. We can now estimate the proportion of reviews that contain each topic by applying string matches on the constituent words. One way to find additional tokens that are relevant to a topic is to use word embeddings. These come from a Word2Vec model trained on our corpus in [this](insert link) notebook."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T19:59:42.983097Z",
     "start_time": "2018-12-17T19:59:40.901431Z"
    }
   },
   "outputs": [],
   "source": [
    "model_big = Word2Vec.load(\"../word2vec_all.model\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T20:06:54.315843Z",
     "start_time": "2018-12-17T20:06:54.310152Z"
    }
   },
   "source": [
    "The `gensim` model has a `most_similar()` function which enables us to retrieve the top n tokens most similar (i.e. closest in n-dimensional space) to one given.\n",
    "\n",
    "For instance, below are the top 10 tokens most similar to 'coffee'. Notice, interestingly, that Word2Vec is an effective way to identify and overcome typos. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T20:09:43.474813Z",
     "start_time": "2018-12-17T20:09:43.467935Z"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[('coffe', 0.8115454912185669),\n",
       " ('espresso', 0.8087444305419922),\n",
       " ('java', 0.7499706745147705),\n",
       " ('coffees', 0.7019534111022949),\n",
       " ('cappuccino', 0.6990188360214233),\n",
       " ('brew', 0.6959743499755859),\n",
       " ('decaf', 0.6938413381576538),\n",
       " ('starbucks', 0.6865791082382202),\n",
       " ('expresso', 0.6716079115867615),\n",
       " ('folgers', 0.6431935429573059)]"
      ]
     },
     "execution_count": 111,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_big.most_similar(positive = ['coffee'], topn=10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can also do sense checks by looking at a sample of reviews that contain a token of interest: "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 142,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T20:27:45.169998Z",
     "start_time": "2018-12-17T20:27:45.124135Z"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "I drink all kinds of non-black decaf teas.  A friend of mine brought some of this Wild Sweet Orange for me to try.  OMG!  Just the smell from the teabag is delicious!  And that incredible taste carries over into the actual brewed tea (unlike coffee, for example, where the beans smell incredible but then you brew them and it's just ..eh.).  It's the smell and flavor of a ripe orange - without any acidity.  It's simply delicious.  Anyway, as soon as I tried it, I bought a box.  And then another.  I love it.\n",
      "\n",
      "This product has good flavor for a decaf coffee.  I was using S&S to keep stocked, but it was canceled.  The only downside to the product is that I had to transfer the pods to a airtight container after opening the bag.  The bag wouldn't open cleanly, and you couldn't get it to close tightly.\n",
      "\n",
      "I am very impressed. I saw the reviews and just had to try some, the reviews for this tea are incredible. The teabags do not have tags or strings, which is a bummer, but the taste of the tea makes you forget about that little detail. It just has this strong, smooth flavor, yet not bitter... This is some of the best tea I've tried. I want to get it in decaf too so I can drink it in the evening without the caffeine keeping me awake. Highly recommend.\n",
      "\n"
     ]
    }
   ],
   "source": [
    "for i in df[df.tokenized_text.str.contains('decaf')].reviewText.values[:3]:\n",
    "    print(i); print('')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Count Topics"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T21:09:59.734756Z",
     "start_time": "2018-12-17T21:09:59.729494Z"
    }
   },
   "source": [
    "We now create a topic dictionary using tokens from the topic models and the Word2Vec model:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 158,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T20:53:42.445813Z",
     "start_time": "2018-12-17T20:53:42.403670Z"
    },
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "coffee_anchor = ['coffee', 'cup', 'drink', 'tea', 'brew', 'keurig', 'kcup', 'green_tea']\n",
    "coffee_extra = [i[0] for i in model_big.most_similar(positive = ['coffee'], topn=10)]\n",
    "ingredients_anchor = ['ingredient', 'calorie', 'fat', 'serving', 'sodium', ' gram', 'contain ', ' per ', 'serve', 'protein']\n",
    "ingredients_extra = [i[0] for i in model_big.most_similar(positive = ['ingredient'], topn=10)]\n",
    "directions_anchor = ['minute', 'prepare', 'microwave', 'boil', 'direction', 'cook', 'easy']\n",
    "directions_extra = [i[0] for i in model_big.most_similar(positive = ['prepare'], topn=10)]\n",
    "price_anchor = ['price', 'save' , 'pay', 'subscribe', 'money', 'cost', 'cheaper', 'reasonably']\n",
    "price_extra = [i[0] for i in model_big.most_similar(positive = ['price'], topn=10)]\n",
    "wheatprods_anchor = ['gluten', 'pasta', 'flour', 'wheat', 'barilla', 'spaghetti', 'bread']\n",
    "wheatprods_extra = [i[0] for i in model_big.most_similar(positive = ['pasta'], topn=10)]\n",
    "sweetener_anchor = ['honey', 'sweetener', 'sweeten', 'stevia']\n",
    "sweetener_extra = [i[0] for i in model_big.most_similar(positive = ['sweetener'], topn=10)]\n",
    "cookies_anchor = ['cooky', 'cookie', 'mix', 'keebler', 'chocolate_chip', 'shortbread', 'pancake']\n",
    "cookies_extra = [i[0] for i in model_big.most_similar(positive = ['cookie'], topn=10)]\n",
    "popcorn_anchor = ['corn', 'popcorn', 'pop', 'kernel', 'orville', 'resealable']\n",
    "popcorn_extra = [i[0] for i in model_big.most_similar(positive = ['popcorn'], topn=10)]\n",
    "oil_anchor = ['oil', 'fat', 'coconut_oil', 'saturate', 'trans', 'cholesterol', 'olive_oil']\n",
    "oil_extra = [i[0] for i in model_big.most_similar(positive = ['olive'], topn=10)]\n",
    "snacks_anchor = ['snack', 'apple', 'crunchy', 'chocolate', 'bar', 'fruit', 'sweet', 'nut', 'cracker']\n",
    "snacks_extra = [i[0] for i in model_big.most_similar(positive = ['snack'], topn=10)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 162,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T21:01:19.303632Z",
     "start_time": "2018-12-17T21:01:19.298957Z"
    }
   },
   "outputs": [],
   "source": [
    "topic_dict = {'t_drinks': coffee_anchor+coffee_extra,\\\n",
    "              't_ingredients': ingredients_anchor+ingredients_extra,\\\n",
    "              't_directions': directions_anchor+directions_extra,\\\n",
    "              't_prices': price_anchor+price_extra,\\\n",
    "              't_pasta_bread': wheatprods_anchor+wheatprods_extra,\\\n",
    "              't_sweetener': sweetener_anchor+sweetener_extra,\\\n",
    "              't_cookies': cookies_anchor+cookies_extra,\\\n",
    "              't_popcorn': popcorn_anchor+popcorn_extra,\\\n",
    "              't_oil': oil_anchor+oil_extra,\\\n",
    "              't_snacks': snacks_anchor+snacks_extra\\\n",
    "             }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 175,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T21:02:18.703309Z",
     "start_time": "2018-12-17T21:02:18.698486Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Topic: t_drinks\n",
      "Words: ['coffee', 'cup', 'drink', 'tea', 'brew', 'keurig', 'kcup', 'green_tea', 'coffe', 'espresso', 'java', 'coffees', 'cappuccino', 'brew', 'decaf', 'starbucks', 'expresso', 'folgers']\n",
      "\n",
      "Topic: t_ingredients\n",
      "Words: ['ingredient', 'calorie', 'fat', 'serving', 'sodium', ' gram', 'contain ', ' per ', 'serve', 'protein', 'ingredients', 'ingrediant', '60why', 'ingrident', 'mid1976', 'paragraph', 'listed', 'component', 'juie', 'maltodextrin']\n",
      "\n",
      "Topic: t_directions\n",
      "Words: ['minute', 'prepare', 'microwave', 'boil', 'direction', 'cook', 'easy', 'cook', 'prep', 'miscalculate', 'access', 'make', 'preparation', 'peasy', 'retrieval', 'overindulge', 'cleanup']\n",
      "\n",
      "Topic: t_prices\n",
      "Words: ['price', 'save', 'pay', 'subscribe', 'money', 'cost', 'cheaper', 'reasonably', 'pricing', 'cost', 'prices', 'priced', 'discount', 'value', 'deal', 'bargain', 'expense', 'hting']\n",
      "\n",
      "Topic: t_pasta_bread\n",
      "Words: ['gluten', 'pasta', 'flour', 'wheat', 'barilla', 'spaghetti', 'bread', 'spaghetti', 'linguine', 'couscous', 'penne', 'rotini', 'lasagna', 'barilla', 'noodles', 'vermicelli', 'pastas']\n",
      "\n",
      "Topic: t_sweetener\n",
      "Words: ['honey', 'sweetener', 'sweeten', 'stevia', 'sweetner', 'sweeteners', 'splenda', 'stevia', 'truvia', 'sucralose', 'sweetening', 'erythritol', 'sweetners', 'nutrasweet']\n",
      "\n",
      "Topic: t_cookies\n",
      "Words: ['cooky', 'cookie', 'mix', 'keebler', 'chocolate_chip', 'shortbread', 'pancake', 'cracker', 'brownie', 'cookies', 'oreo', 'bar', 'marshmallow', 'wafer', 'fudge', 'shortbread', 'crust']\n",
      "\n",
      "Topic: t_popcorn\n",
      "Words: ['corn', 'popcorn', 'pop', 'kernel', 'orville', 'resealable', 'popper', 'orville', 'redenbacher', 'theater', 'movie', 'popcorns', 'cheddar', 'stuff', 'kernels', 'popped']\n",
      "\n",
      "Topic: t_oil\n",
      "Words: ['oil', 'fat', 'coconut_oil', 'saturate', 'trans', 'cholesterol', 'olive_oil', 'grapeseed', 'vegeatables', 'avocado', 'soybean', 'aniseed', 'basalmic', 'ouncestaste', 'coconut', 'canola', 'safflower']\n",
      "\n",
      "Topic: t_snacks\n",
      "Words: ['snack', 'apple', 'crunchy', 'chocolate', 'bar', 'fruit', 'sweet', 'nut', 'cracker', 'snacks', 'treat', 'snacking', 'lunch', 'pickmeup', 'meal', 'dessert', 'onthego', 'lunches', 'treats']\n",
      "\n"
     ]
    }
   ],
   "source": [
    "for i in topic_dict:\n",
    "    print('Topic:', i)\n",
    "    print('Words:', topic_dict[i])\n",
    "    print('')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 181,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T21:11:37.580065Z",
     "start_time": "2018-12-17T21:11:34.366474Z"
    },
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "t_drinks         33.86\n",
       "t_ingredients    22.55\n",
       "t_directions     40.52\n",
       "t_prices         23.74\n",
       "t_pasta_bread    13.57\n",
       "t_sweetener       6.89\n",
       "t_cookies        21.80\n",
       "t_popcorn        12.70\n",
       "t_oil            14.55\n",
       "t_snacks         47.33\n",
       "dtype: float64"
      ]
     },
     "execution_count": 181,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_matches = df.copy()\n",
    "\n",
    "for key, value in topic_dict.items():\n",
    "    word_list = '|'.join(value)\n",
    "    df_matches[key] = df_matches.tokenized_text.str.contains(word_list).astype(int)\n",
    "    \n",
    "df_matches[df_matches.columns[-len(topic_dict.keys()):]].sum(axis = 0).apply(lambda x: round(x/len(df_matches)*100,2))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-14T18:23:55.666626Z",
     "start_time": "2018-12-14T18:23:55.638553Z"
    }
   },
   "source": [
    "The output shows that ~34% of reviews are coffee/tea related and ~50% of the reviews are snack related. 24% of reviews discuss cost, etc."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 186,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T21:22:02.425693Z",
     "start_time": "2018-12-17T21:22:02.408684Z"
    },
    "scrolled": false
   },
   "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>reviewerID</th>\n",
       "      <th>asin</th>\n",
       "      <th>reviewerName</th>\n",
       "      <th>helpful</th>\n",
       "      <th>reviewText</th>\n",
       "      <th>overall</th>\n",
       "      <th>summary</th>\n",
       "      <th>unixReviewTime</th>\n",
       "      <th>reviewTime</th>\n",
       "      <th>review_year</th>\n",
       "      <th>reviews_clean</th>\n",
       "      <th>reviews_regram</th>\n",
       "      <th>tokenized_text</th>\n",
       "      <th>t_drinks</th>\n",
       "      <th>t_ingredients</th>\n",
       "      <th>t_directions</th>\n",
       "      <th>t_prices</th>\n",
       "      <th>t_pasta_bread</th>\n",
       "      <th>t_sweetener</th>\n",
       "      <th>t_cookies</th>\n",
       "      <th>t_popcorn</th>\n",
       "      <th>t_oil</th>\n",
       "      <th>t_snacks</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>A1VEELTKS8NLZB</td>\n",
       "      <td>616719923X</td>\n",
       "      <td>Amazon Customer</td>\n",
       "      <td>[0, 0]</td>\n",
       "      <td>Just another flavor of Kit Kat but the taste i...</td>\n",
       "      <td>4.0</td>\n",
       "      <td>Good Taste</td>\n",
       "      <td>1370044800</td>\n",
       "      <td>06 1, 2013</td>\n",
       "      <td>2013</td>\n",
       "      <td>just another flavor of kit kat but the taste i...</td>\n",
       "      <td>just another flavor of kit kat but the taste i...</td>\n",
       "      <td>another flavor kit kat taste unique bit differ...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>A27IQHDZFQFNGG</td>\n",
       "      <td>616719923X</td>\n",
       "      <td>Caitlin</td>\n",
       "      <td>[3, 4]</td>\n",
       "      <td>Really good. Great gift for any fan of green t...</td>\n",
       "      <td>4.0</td>\n",
       "      <td>Yum!</td>\n",
       "      <td>1381190400</td>\n",
       "      <td>10 8, 2013</td>\n",
       "      <td>2013</td>\n",
       "      <td>really good  great gift for any fan of green t...</td>\n",
       "      <td>really good  great gift for any fan of green_t...</td>\n",
       "      <td>really good great gift fan green_tea expensive...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       reviewerID        asin     reviewerName helpful  \\\n",
       "0  A1VEELTKS8NLZB  616719923X  Amazon Customer  [0, 0]   \n",
       "1  A27IQHDZFQFNGG  616719923X          Caitlin  [3, 4]   \n",
       "\n",
       "                                          reviewText  overall     summary  \\\n",
       "0  Just another flavor of Kit Kat but the taste i...      4.0  Good Taste   \n",
       "1  Really good. Great gift for any fan of green t...      4.0        Yum!   \n",
       "\n",
       "   unixReviewTime  reviewTime  review_year  \\\n",
       "0      1370044800  06 1, 2013         2013   \n",
       "1      1381190400  10 8, 2013         2013   \n",
       "\n",
       "                                       reviews_clean  \\\n",
       "0  just another flavor of kit kat but the taste i...   \n",
       "1  really good  great gift for any fan of green t...   \n",
       "\n",
       "                                      reviews_regram  \\\n",
       "0  just another flavor of kit kat but the taste i...   \n",
       "1  really good  great gift for any fan of green_t...   \n",
       "\n",
       "                                      tokenized_text  t_drinks  t_ingredients  \\\n",
       "0  another flavor kit kat taste unique bit differ...         0              0   \n",
       "1  really good great gift fan green_tea expensive...         1              0   \n",
       "\n",
       "   t_directions  t_prices  t_pasta_bread  t_sweetener  t_cookies  t_popcorn  \\\n",
       "0             0         1              0            0          0          0   \n",
       "1             0         0              0            0          0          0   \n",
       "\n",
       "   t_oil  t_snacks  \n",
       "0      0         0  \n",
       "1      0         0  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "display_all(df_matches[:2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 183,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T21:17:49.099183Z",
     "start_time": "2018-12-17T21:17:47.031843Z"
    }
   },
   "outputs": [],
   "source": [
    "now = time.strftime('%Y%m%d_%H%M')\n",
    "df_matches.to_csv('tm_output_%s.csv' %now, index = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And that's it! Note that this is just a toy example and additional tuning that can be applied include:\n",
    "- a wider grid search for hyperparameter `n` in the topic models \n",
    "- anchoring pre-identified words of interest. For instance, if my client was Heinz and they wanted to understand what proportion of reviews relate to sauces, I could seed a topic constituting of anchor words around 'ketchup', 'mustard', 'mayonaise', and similar words identified through the Word2Vec model.\n",
    "- more sophistcaed word matching logic. For instance, if I was particular interested in 'salty popcorn', I would not use the syntax cabove, which would catch all reviews containg 'salty' and 'popcorn' separately. I would also not want to match on 'salty popcorn' itself, which would miss reviews that say, 'This popcorn is too salty'. Instead, I could use something like this: "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-12-17T21:59:00.118433Z",
     "start_time": "2018-12-17T21:59:00.109349Z"
    },
    "scrolled": false
   },
   "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>flag</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>this popcorn is too sweet, do not recommend</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>gosh this seasalt is salty</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>this popcorn contains 100mg salt</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>i could live on salty popcorn</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                             0  flag\n",
       "0  this popcorn is too sweet, do not recommend     0\n",
       "1                   gosh this seasalt is salty     0\n",
       "2             this popcorn contains 100mg salt     1\n",
       "3                i could live on salty popcorn     1"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "word_ls = ['salt', 'popcorn']\n",
    "sentence1 = 'this popcorn is too sweet, do not recommend'\n",
    "sentence2 = 'gosh this seasalt is salty'\n",
    "sentence3 = 'this popcorn contains 100mg salt'\n",
    "sentence4 = 'i could live on salty popcorn'\n",
    "test = pd.DataFrame(list([sentence1, sentence2, sentence3, sentence4]))\n",
    "test['flag'] = test[0].apply(lambda x: int(all(word in x for word in word_ls)))\n",
    "test"
   ]
  }
 ],
 "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": "3.6.5"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {
    "height": "608px",
    "left": "545px",
    "top": "110px",
    "width": "259px"
   },
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}