{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "# Sentiment classification with VADER"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## loading IMDB dataset\n",
    "Only the test data is loaded, since VADER does not require training data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#download only once\n",
    "from urllib import request\n",
    "url = \"https://goo.gl/mg8bsD\"\n",
    "response = request.urlopen(url)\n",
    "text = response.read().decode('utf-8')\n",
    "with open('imdb_test.txt',mode='w',encoding='utf-8') as outputfile:\n",
    "    outputfile.write(text)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('imdb_test.txt',mode='r',encoding='utf-8') as inputfile:\n",
    "    text = inputfile.read()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import csv\n",
    "x_test = list()\n",
    "y_test = list()\n",
    "with open('imdb_test.txt', encoding='utf-8', newline='') as infile:\n",
    "    reader = csv.reader(infile, delimiter='\\t')\n",
    "    for row in reader:\n",
    "        x_test.append(row[0])\n",
    "        y_test.append(int(row[1]))\n",
    "x_test[0],y_test[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "len(x_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## creating VADER classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from nltk.sentiment.vader import SentimentIntensityAnalyzer\n",
    "vader = SentimentIntensityAnalyzer()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "vader.polarity_scores('not the best experience I had')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "vader.polarity_scores(':P')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## classification of test data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "scores = list()\n",
    "for text in x_test:\n",
    "    scores.append(vader.polarity_scores(text)['compound'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "list(zip(scores,y_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's try to assign labels only when the confidence is very high.\n",
    "This is not useful when you must classify all documents.\n",
    "It is instead useful when you want to bootstrap a training set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "selection = []\n",
    "for score,label in list(zip(scores,y_test)):\n",
    "    if abs(score)>0.99:\n",
    "        selection.append((score,label))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "len(selection),len(y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Evaluation of accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "accuracy = 0\n",
    "for prediction,correct in zip(scores, y_test):\n",
    "    if prediction>0 and correct==1 or prediction<=0 and correct==0:\n",
    "        accuracy += 1\n",
    "print(len(scores),accuracy/len(scores))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following evaluation is not fair, because it is not computed on the full training set.\n",
    "Yet it shows that the subset of documents that get a label is more accurately labeled.\n",
    "So if such subset is used as a training set for a supervised learning algorithm you can expect to learn a better classifier."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "accuracy = 0\n",
    "for prediction,correct in selection:\n",
    "    if prediction>0 and correct==1 or prediction<=0 and correct==0:\n",
    "        accuracy += 1\n",
    "print(len(selection),accuracy/len(selection))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (TA)",
   "language": "python",
   "name": "ta"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}