yelp-review-classifier.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Yelp Review Classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# https://medium.com/@ageitgey/text-classification-is-your-new-secret-weapon-7ca4fad15788\n",
    "# data: https://www.yelp.com/dataset/download\n",
    "import json\n",
    "from pathlib import Path\n",
    "import re\n",
    "import random\n",
    "import fastText"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "#PRE PROCESS, Only run to make/clean new data sets\n",
    "reviews_data = Path(\"dataset\") / \"review.json\"\n",
    "training_data = Path(\"fasttext_dataset_training.txt\")\n",
    "test_data = Path(\"fasttext_dataset_test.txt\")\n",
    "\n",
    "# What percent of data to save separately as test data\n",
    "percent_test_data = 0.10\n",
    "\n",
    "def strip_formatting(string):\n",
    "    string = string.lower()\n",
    "    string = re.sub(r\"([.!?,'/()])\", r\" \\1 \", string)\n",
    "    return string\n",
    "\n",
    "with reviews_data.open() as input, \\\n",
    "     training_data.open(\"w\") as train_output, \\\n",
    "     test_data.open(\"w\") as test_output:\n",
    "\n",
    "    for line in input:\n",
    "        review_data = json.loads(line)\n",
    "\n",
    "        rating = review_data['stars']\n",
    "        text = review_data['text'].replace(\"\\n\", \" \")\n",
    "        text = strip_formatting(text)\n",
    "\n",
    "        fasttext_line = \"__label__{} {}\".format(rating, text)\n",
    "\n",
    "        if random.random() <= percent_test_data:\n",
    "            test_output.write(fasttext_line + \"\\n\")\n",
    "        else:\n",
    "            train_output.write(fasttext_line + \"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def strip_formatting(string):\n",
    "    string = string.lower()\n",
    "    string = re.sub(r\"([.!?,'/()])\", r\" \\1 \", string)\n",
    "    return string"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Reviews to check\n",
    "reviews = [\n",
    "    \"This restaurant literally changed my life. This is the best food I've ever eaten!\",\n",
    "    \"I hate this place so much. They were mean to me.\",\n",
    "    \"I don't know. It was ok, I guess. Not really sure what to say.\"\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Pre-process the text of each review so it matches the training format\n",
    "preprocessed_reviews = list(map(strip_formatting, reviews))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load the model\n",
    "classifier = fastText.load_model('../../../fasttext/reviews_model_ngrams.bin')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Get fastText to classify each review with the model\n",
    "labels, probabilities = classifier.predict(preprocessed_reviews, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " (100% confidence)\n",
      "This restaurant literally changed my life. This is the best food I've ever eaten!\n",
      "\n",
      " (97% confidence)\n",
      "I hate this place so much. They were mean to me.\n",
      "\n",
      " (83% confidence)\n",
      "I don't know. It was ok, I guess. Not really sure what to say.\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Print the results\n",
    "for review, label, probability in zip(reviews, labels, probabilities):\n",
    "    stars = int(label[0][-1])\n",
    "\n",
    "    print(\"{} ({}% confidence)\".format(\"☆\" * stars, int(probability[0] * 100)))\n",
    "    print(review)\n",
    "    print()"
   ]
  }
 ],
 "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}