simple_logisitc_regression.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Predicting Amazon review ratings based only on the text using TF-IDF and Logistic Regression\n",
    " \n",
    "In this exercise I will use a simple Logistic Regression model to explore if we can predict the rating of a Book review in Amazon based exclusively on the text of the review.\n",
    " \n",
    "I will begin importing the libraries necessary to perform this exercise."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[nltk_data] Downloading package stopwords to\n",
      "[nltk_data]     /home/ordovas/nltk_data...\n",
      "[nltk_data]   Package stopwords is already up-to-date!\n",
      "[nltk_data] Downloading package wordnet to /home/ordovas/nltk_data...\n",
      "[nltk_data]   Package wordnet is already up-to-date!\n",
      "[nltk_data] Downloading package punkt to /home/ordovas/nltk_data...\n",
      "[nltk_data]   Package punkt is already up-to-date!\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from pymongo import MongoClient\n",
    "\n",
    "import re\n",
    "\n",
    "import nltk\n",
    "from nltk.corpus import stopwords\n",
    "nltk.download(\"stopwords\")\n",
    "from nltk.stem import WordNetLemmatizer\n",
    "nltk.download('wordnet')\n",
    "nltk.download('punkt')\n",
    "\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.metrics import confusion_matrix\n",
    "from sklearn.metrics import plot_confusion_matrix\n",
    "\n",
    "client = MongoClient()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we load a fraction of the dataset and visualize the head of the dataset that we obtained."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "db = client.get_database(\"amazon\")\n",
    "# Define an alias for the books dataset\n",
    "books =db.books\n",
    "# Obtain a random sample from the dataset, selecting only a few records \n",
    "# (we will use only overall and reviewTest, but I will load a few more just in case I\n",
    "# want to play with more info...)\n",
    "res = list(books.aggregate([\n",
    "    { \"$sample\": { \"size\": 75000 }}\n",
    "    ,{ \"$project\": {\"id\": \"$_id\", \"_id\": 0, \"overall\": 1, \"reviewText\": 1,\"summary\":1,\"reviewerName\":1}} \n",
    "]))\n",
    "# Converting to pandas DataFrame\n",
    "df=pd.DataFrame(res)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "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>overall</th>\n",
       "      <th>reviewerName</th>\n",
       "      <th>reviewText</th>\n",
       "      <th>summary</th>\n",
       "      <th>id</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2</td>\n",
       "      <td>Lauren Page</td>\n",
       "      <td>This book was just horrid. I found myself just...</td>\n",
       "      <td>Nothing like the other books. Just bad.</td>\n",
       "      <td>6033ab76f7ccad4a92f27666</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>Taka</td>\n",
       "      <td>In the first twenty or thirty pages, I was ple...</td>\n",
       "      <td>OVERrated</td>\n",
       "      <td>6033ab85f7ccad4a92f55229</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>5</td>\n",
       "      <td>Samuel J.</td>\n",
       "      <td>Obviously the story is a classic for a reason,...</td>\n",
       "      <td>Very Nice</td>\n",
       "      <td>6033ab79f7ccad4a92f30a94</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>5</td>\n",
       "      <td>Mary M. Anderson</td>\n",
       "      <td>I wish I had seen this book when I was first b...</td>\n",
       "      <td>Best Book Ever</td>\n",
       "      <td>6033ab7df7ccad4a92f3c7df</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5</td>\n",
       "      <td>DiElizabeth</td>\n",
       "      <td>This is a must read book , especially before w...</td>\n",
       "      <td>Magical adventure</td>\n",
       "      <td>6033ab79f7ccad4a92f307e5</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   overall      reviewerName  \\\n",
       "0        2       Lauren Page   \n",
       "1        2              Taka   \n",
       "2        5         Samuel J.   \n",
       "3        5  Mary M. Anderson   \n",
       "4        5       DiElizabeth   \n",
       "\n",
       "                                          reviewText  \\\n",
       "0  This book was just horrid. I found myself just...   \n",
       "1  In the first twenty or thirty pages, I was ple...   \n",
       "2  Obviously the story is a classic for a reason,...   \n",
       "3  I wish I had seen this book when I was first b...   \n",
       "4  This is a must read book , especially before w...   \n",
       "\n",
       "                                   summary                        id  \n",
       "0  Nothing like the other books. Just bad.  6033ab76f7ccad4a92f27666  \n",
       "1                                OVERrated  6033ab85f7ccad4a92f55229  \n",
       "2                                Very Nice  6033ab79f7ccad4a92f30a94  \n",
       "3                           Best Book Ever  6033ab7df7ccad4a92f3c7df  \n",
       "4                        Magical adventure  6033ab79f7ccad4a92f307e5  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "I wil now drop the rows with NaN values in the `reviewText` column."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 75000 entries, 0 to 74999\n",
      "Data columns (total 5 columns):\n",
      " #   Column        Non-Null Count  Dtype \n",
      "---  ------        --------------  ----- \n",
      " 0   overall       75000 non-null  int64 \n",
      " 1   reviewerName  75000 non-null  object\n",
      " 2   reviewText    74989 non-null  object\n",
      " 3   summary       74989 non-null  object\n",
      " 4   id            75000 non-null  object\n",
      "dtypes: int64(1), object(4)\n",
      "memory usage: 2.9+ MB\n"
     ]
    }
   ],
   "source": [
    "df.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "df=df.dropna(subset=['reviewText'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's check that everything is fine..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 74989 entries, 0 to 74999\n",
      "Data columns (total 5 columns):\n",
      " #   Column        Non-Null Count  Dtype \n",
      "---  ------        --------------  ----- \n",
      " 0   overall       74989 non-null  int64 \n",
      " 1   reviewerName  74989 non-null  object\n",
      " 2   reviewText    74989 non-null  object\n",
      " 3   summary       74978 non-null  object\n",
      " 4   id            74989 non-null  object\n",
      "dtypes: int64(1), object(4)\n",
      "memory usage: 3.4+ MB\n"
     ]
    }
   ],
   "source": [
    "df.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "OK, so now I will examine the `overall` column, that is the target feature.\n",
    " \n",
    "There are far more reviews with high ratings than with low scores, so there is no balance. To solve this, later on I will weight the clases instead of selecting now the exact number of reviews for each rating."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "5    47257\n",
       "4    15151\n",
       "3     6715\n",
       "2     3152\n",
       "1     2714\n",
       "Name: overall, dtype: int64"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df[\"overall\"].value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Preprocessing: Cleaning text and TF-IDF vectorization\n",
    " \n",
    " \n",
    "To begin with I will clean the review text ignoring stop words and lemmatizing all the words."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to lemmatize the text of the review\n",
    "def lemmatizer_text(text):\n",
    "    # Load Lemmatizer\n",
    "    lemmatizer=WordNetLemmatizer()\n",
    "    # Obtain a list with each word\n",
    "    words = nltk.word_tokenize(text)\n",
    "    # Returns an string joining the lemmatized words\n",
    "    return ' '.join([lemmatizer.lemmatize(w) for w in words])\n",
    "\n",
    "\n",
    "def cleaning_review(texts):\n",
    "    #Remove numbers, punctuation and lowercase everything\n",
    "    res=re.sub(\"[^A-Za-z]+\", \" \", texts.lower()) \n",
    "    #Lemmanize the text of the review\n",
    "    res=lemmatizer_text(res)\n",
    "    #Remove words with less than 4 letters\n",
    "    res=re.sub(r'\\b\\w{1,3}\\b', \"\", res) \n",
    "    #Remove rest of stop words\n",
    "    for stopword in stopwords.words('english'): \n",
    "        sw=stopword.replace(\"'\",\"\")\n",
    "        res=res.replace(f\" {sw} \",\" \")\n",
    "    return res\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now I will create a feature with the \"cleaned\" text named `review_clean`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "df[\"review_clean\"]=df[\"reviewText\"]\n",
    "df[\"review_clean\"]=df[\"review_clean\"].apply(cleaning_review)\n",
    "df=df.reset_index()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And then we can now create the TF-IDF vectors for each review using `TfidfVectorizer` from  `sklearn`. We ignore words that don't appear in at least 5 reviews.\n",
    "\n",
    "But first, we need to split the dataset into train and test samples. I will choose 85% of the sample to train and 15% to test. The TF-IDF vectorizer needs to be trained first on the train dataset and after that applied to the test dataset. This is done in order to prevent data leakage, that means that information outside of the train dataset is used during the trainin phase. The TF-IDF vectorizer is applied to the test dataset *after* it is trained with the train dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "Xraw_train, Xraw_test, y_train, y_test = train_test_split(df[\"review_clean\"], df[\"overall\"], test_size=0.15, random_state=42)\n",
    "\n",
    "v = TfidfVectorizer(min_df=5)\n",
    "tf_idf_train = v.fit_transform(Xraw_train).toarray()\n",
    "tf_idf_test = v.transform(Xraw_test).toarray()\n",
    "\n",
    "X_train=pd.DataFrame(tf_idf_train, columns=v.get_feature_names())\n",
    "X_test=pd.DataFrame(tf_idf_test, columns=v.get_feature_names())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Just for curiosity, let's check how many features/words the TF-IDF vectors have..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "20476"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(v.get_feature_names())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And let's check the size of the train and test datasets."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(63740, 11249)"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(y_train),len(y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Logistic Regression: Predicting the scores based on the TF-IDF vectors\n",
    "\n",
    "Now it's time to use the Logistic Regression model to predict the scores of the reviews. The TD-IFD vectors are the features used to predict the target, that is the `overall` column that stores the numeric ratings.\n",
    " \n",
    "We will activate `class_weight=\"balanced\"`, as the classes (scores) are not balanced, so if we do not weigh the classes the fit will be biased towards high ratings (most populated classes).\n",
    " \n",
    "This is just a very simple approach, without fine tuning anything. This is just an exercise for learning purposes."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ordovas/anaconda3/envs/ironhack/lib/python3.8/site-packages/sklearn/linear_model/_logistic.py:762: ConvergenceWarning: lbfgs failed to converge (status=1):\n",
      "STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.\n",
      "\n",
      "Increase the number of iterations (max_iter) or scale the data as shown in:\n",
      "    https://scikit-learn.org/stable/modules/preprocessing.html\n",
      "Please also refer to the documentation for alternative solver options:\n",
      "    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression\n",
      "  n_iter_i = _check_optimize_result(\n"
     ]
    }
   ],
   "source": [
    "clf = LogisticRegression(random_state=0,class_weight=\"balanced\").fit(X_train, y_train)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "That's it? Well, now we need to check the performance of this model...\n",
    "\n",
    "Let's check the score of the classification model for the train and test samples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train sample classification score: 0.68\n",
      "Test sample classification score: 0.57\n"
     ]
    }
   ],
   "source": [
    "sc_train=clf.score(X_train, y_train)\n",
    "sc_test=clf.score(X_test, y_test)\n",
    "print(f\"Train sample classification score: {round(sc_train,2)}\")\n",
    "print(f\"Test sample classification score: {round(sc_test,2)}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This seems just OK, but let's check an histogram to see the accuracy of the predictions by plotting the true score minus the predicted score."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "counts, bins = np.histogram(y_train-clf.predict(X_train),bins=np.arange(-4,6))\n",
    "plt.plot(bins[:-1],counts*100/sum(counts),label=\"Train data\")\n",
    "plt.grid()\n",
    "counts, bins = np.histogram(y_test-clf.predict(X_test),bins=np.arange(-4,6))\n",
    "plt.plot(bins[:-1],counts*100/sum(counts),label=\"Test data\")\n",
    "plt.xlabel(\"Real Rating - Predicted Rating\")\n",
    "plt.ylabel(\"Percentage\")\n",
    "plt.legend();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Well, this simple model seems not to be very wrong! Let's quantify this with the confusion matrices as well."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 2082,   109,    45,    23,    53],\n",
       "       [  139,  2306,   143,    56,    55],\n",
       "       [  331,   555,  3738,   626,   430],\n",
       "       [  395,   596,  1708,  7495,  2690],\n",
       "       [ 1250,  1123,  2456,  7509, 27827]])"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "confusion_matrix(y_train,clf.predict(X_train))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 195,  114,   40,   18,   35],\n",
       "       [ 114,  180,   96,   36,   27],\n",
       "       [ 101,  198,  412,  220,  104],\n",
       "       [  88,  138,  434,  957,  650],\n",
       "       [ 253,  223,  444, 1449, 4723]])"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "confusion_matrix(y_test,clf.predict(X_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And normalized to the true scores."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train subsample:\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test subsample:\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(\"Train subsample:\")\n",
    "plot_confusion_matrix(clf, X_train, y_train,cmap=\"viridis\",normalize=\"true\") \n",
    "plt.show();\n",
    "print(\"Test subsample:\")\n",
    "plot_confusion_matrix(clf, X_test, y_test,cmap=\"viridis\",normalize=\"true\")\n",
    "plt.show();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The diagonal indicates that the Logistic Model predicts reasonably well the score of the review for this simple approach."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Undersanding the Logistic Regression: How it is pedicted\n",
    "\n",
    "I will create a DataFrame with the coefficients of each score to understand how the features (words) contributes to each class."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "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>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>aaron</th>\n",
       "      <td>-0.061693</td>\n",
       "      <td>0.192707</td>\n",
       "      <td>-0.142920</td>\n",
       "      <td>-0.162202</td>\n",
       "      <td>0.174108</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>aback</th>\n",
       "      <td>-0.181668</td>\n",
       "      <td>0.242306</td>\n",
       "      <td>-0.086103</td>\n",
       "      <td>-0.003963</td>\n",
       "      <td>0.029428</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abandon</th>\n",
       "      <td>0.897491</td>\n",
       "      <td>-0.205578</td>\n",
       "      <td>-0.376557</td>\n",
       "      <td>-0.203174</td>\n",
       "      <td>-0.112182</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abandoned</th>\n",
       "      <td>0.836370</td>\n",
       "      <td>-0.211472</td>\n",
       "      <td>-0.313179</td>\n",
       "      <td>0.016732</td>\n",
       "      <td>-0.328450</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abandoning</th>\n",
       "      <td>0.149999</td>\n",
       "      <td>-0.252986</td>\n",
       "      <td>-0.019347</td>\n",
       "      <td>0.093836</td>\n",
       "      <td>0.028497</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abandonment</th>\n",
       "      <td>0.057166</td>\n",
       "      <td>0.039397</td>\n",
       "      <td>0.150065</td>\n",
       "      <td>-0.095506</td>\n",
       "      <td>-0.151121</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abarat</th>\n",
       "      <td>-0.578782</td>\n",
       "      <td>0.478916</td>\n",
       "      <td>-0.028714</td>\n",
       "      <td>-0.121703</td>\n",
       "      <td>0.250284</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abbey</th>\n",
       "      <td>-0.394625</td>\n",
       "      <td>0.721646</td>\n",
       "      <td>-0.037053</td>\n",
       "      <td>-0.662660</td>\n",
       "      <td>0.372692</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abbot</th>\n",
       "      <td>-0.078685</td>\n",
       "      <td>0.149829</td>\n",
       "      <td>-0.064281</td>\n",
       "      <td>0.049720</td>\n",
       "      <td>-0.056583</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abbott</th>\n",
       "      <td>-0.260464</td>\n",
       "      <td>0.379576</td>\n",
       "      <td>0.055833</td>\n",
       "      <td>0.048523</td>\n",
       "      <td>-0.223469</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abbreviated</th>\n",
       "      <td>-0.543059</td>\n",
       "      <td>1.686415</td>\n",
       "      <td>-0.289327</td>\n",
       "      <td>-0.558401</td>\n",
       "      <td>-0.295627</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abbreviation</th>\n",
       "      <td>-0.024601</td>\n",
       "      <td>-0.061625</td>\n",
       "      <td>0.027105</td>\n",
       "      <td>0.001615</td>\n",
       "      <td>0.057506</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abby</th>\n",
       "      <td>-0.110741</td>\n",
       "      <td>-0.171615</td>\n",
       "      <td>-0.112133</td>\n",
       "      <td>0.539571</td>\n",
       "      <td>-0.145083</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abduct</th>\n",
       "      <td>0.094952</td>\n",
       "      <td>-0.081569</td>\n",
       "      <td>0.006955</td>\n",
       "      <td>0.018338</td>\n",
       "      <td>-0.038676</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abducted</th>\n",
       "      <td>-0.151360</td>\n",
       "      <td>-0.229559</td>\n",
       "      <td>0.324450</td>\n",
       "      <td>0.035106</td>\n",
       "      <td>0.021364</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abduction</th>\n",
       "      <td>-0.130072</td>\n",
       "      <td>0.588694</td>\n",
       "      <td>-0.349617</td>\n",
       "      <td>0.089947</td>\n",
       "      <td>-0.198951</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abel</th>\n",
       "      <td>0.436250</td>\n",
       "      <td>-0.254200</td>\n",
       "      <td>0.120044</td>\n",
       "      <td>-0.188493</td>\n",
       "      <td>-0.113601</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>abercrombie</th>\n",
       "      <td>-0.299208</td>\n",
       "      <td>-0.222743</td>\n",
       "      <td>-0.037001</td>\n",
       "      <td>0.263565</td>\n",
       "      <td>0.295387</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>aberdeen</th>\n",
       "      <td>0.365583</td>\n",
       "      <td>-0.634535</td>\n",
       "      <td>0.080342</td>\n",
       "      <td>0.448670</td>\n",
       "      <td>-0.260061</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>aberration</th>\n",
       "      <td>0.169974</td>\n",
       "      <td>0.533285</td>\n",
       "      <td>-0.186495</td>\n",
       "      <td>-0.278019</td>\n",
       "      <td>-0.238745</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                     1         2         3         4         5\n",
       "aaron        -0.061693  0.192707 -0.142920 -0.162202  0.174108\n",
       "aback        -0.181668  0.242306 -0.086103 -0.003963  0.029428\n",
       "abandon       0.897491 -0.205578 -0.376557 -0.203174 -0.112182\n",
       "abandoned     0.836370 -0.211472 -0.313179  0.016732 -0.328450\n",
       "abandoning    0.149999 -0.252986 -0.019347  0.093836  0.028497\n",
       "abandonment   0.057166  0.039397  0.150065 -0.095506 -0.151121\n",
       "abarat       -0.578782  0.478916 -0.028714 -0.121703  0.250284\n",
       "abbey        -0.394625  0.721646 -0.037053 -0.662660  0.372692\n",
       "abbot        -0.078685  0.149829 -0.064281  0.049720 -0.056583\n",
       "abbott       -0.260464  0.379576  0.055833  0.048523 -0.223469\n",
       "abbreviated  -0.543059  1.686415 -0.289327 -0.558401 -0.295627\n",
       "abbreviation -0.024601 -0.061625  0.027105  0.001615  0.057506\n",
       "abby         -0.110741 -0.171615 -0.112133  0.539571 -0.145083\n",
       "abduct        0.094952 -0.081569  0.006955  0.018338 -0.038676\n",
       "abducted     -0.151360 -0.229559  0.324450  0.035106  0.021364\n",
       "abduction    -0.130072  0.588694 -0.349617  0.089947 -0.198951\n",
       "abel          0.436250 -0.254200  0.120044 -0.188493 -0.113601\n",
       "abercrombie  -0.299208 -0.222743 -0.037001  0.263565  0.295387\n",
       "aberdeen      0.365583 -0.634535  0.080342  0.448670 -0.260061\n",
       "aberration    0.169974  0.533285 -0.186495 -0.278019 -0.238745"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model=pd.DataFrame(clf.coef_,columns=X_train.columns)\n",
    "model=model.T\n",
    "model.columns=[\"1\",\"2\",\"3\",\"4\",\"5\"]\n",
    "model.head(20)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For the highest score, the words that are directly correlated and contributes the most are words as \"excellent\", \"great\", \"love\", \"fantastic\", \"amazing\"..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "excellent      4.601775\n",
       "wonderful      4.093400\n",
       "amazing        3.805142\n",
       "highly         3.804939\n",
       "love           3.625317\n",
       "loved          3.539603\n",
       "perfect        3.433680\n",
       "fantastic      3.384947\n",
       "beautiful      3.282861\n",
       "beautifully    3.177139\n",
       "awesome        3.147184\n",
       "great          3.030050\n",
       "today          3.019786\n",
       "provoking      2.910257\n",
       "terrific       2.890560\n",
       "thank          2.855471\n",
       "wait           2.746846\n",
       "masterpiece    2.656303\n",
       "outstanding    2.606295\n",
       "powerful       2.586116\n",
       "Name: 5, dtype: float64"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model[\"5\"].sort_values(ascending=False).head(20)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So, for the lowest score, it is not strange that the words that are directly correlated and contribute the most are words such as \"terrible\",\"trash\" or \"boring\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "waste             6.413316\n",
       "wasted            5.712731\n",
       "worst             5.666858\n",
       "disappointment    4.118679\n",
       "terrible          4.072798\n",
       "boring            4.052279\n",
       "poorly            3.826763\n",
       "hated             3.715275\n",
       "stupid            3.712801\n",
       "nonsense          3.654662\n",
       "trash             3.597680\n",
       "amazon            3.431366\n",
       "zero              3.413424\n",
       "garbage           3.410193\n",
       "horrible          3.402571\n",
       "crap              3.393276\n",
       "junk              3.380922\n",
       "rubbish           3.298719\n",
       "porn              3.290351\n",
       "apparently        3.265722\n",
       "Name: 1, dtype: float64"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model[\"1\"].sort_values(ascending=False).head(20)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "With respect to the intermediate scores, the words that define these reviews are related to texts that show good and bad aspects of the product, so words like \"however\" or \"unfortunately\" are weighted high, as well as \"okay\",\"somewhat\" and \"alright\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "however         3.674379\n",
       "okay            3.387344\n",
       "frustrating     2.719719\n",
       "overall         2.685825\n",
       "three           2.355088\n",
       "turner          2.252566\n",
       "decent          2.237369\n",
       "seemed          2.171566\n",
       "blown           2.124443\n",
       "felt            2.119106\n",
       "alright         2.041549\n",
       "still           2.032812\n",
       "setting         1.976933\n",
       "particularly    1.973048\n",
       "average         1.972463\n",
       "unusual         1.950792\n",
       "fairly          1.939594\n",
       "lacking         1.922384\n",
       "pleasant        1.914850\n",
       "prefer          1.909206\n",
       "Name: 3, dtype: float64"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model[\"3\"].sort_values(ascending=False).head(20)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "On the other side, words with the lowest value of the coefficients show other information: features that are least likely to appear on a review. For instance, for the best reviewed books, we will not find words as \"boring\" or \"disappointing\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "disappointing    -5.261526\n",
       "boring           -5.004400\n",
       "unfortunately    -4.079230\n",
       "however          -3.951741\n",
       "tedious          -3.859695\n",
       "instead          -3.620947\n",
       "disappointed     -3.578610\n",
       "waste            -3.477536\n",
       "disappointment   -3.290387\n",
       "okay             -3.273860\n",
       "slow             -3.200287\n",
       "seemed           -3.158649\n",
       "predictable      -3.127049\n",
       "poorly           -3.116196\n",
       "depressing       -3.054896\n",
       "annoying         -2.966359\n",
       "premise          -2.946118\n",
       "confusing        -2.853311\n",
       "would            -2.812168\n",
       "difficult        -2.782072\n",
       "Name: 5, dtype: float64"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model[\"5\"].sort_values(ascending=True).head(20)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For the products with lowest score, we will not find \"perfect\", \"good\", \"enjoyed\" or \"great\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "easy           -4.638756\n",
       "beautiful      -3.589769\n",
       "particularly   -3.404560\n",
       "enjoyed        -3.359470\n",
       "excellent      -3.342778\n",
       "turner         -3.303818\n",
       "perspective    -3.113625\n",
       "today          -3.100802\n",
       "fascinating    -3.099424\n",
       "still          -2.954108\n",
       "especially     -2.874138\n",
       "expect         -2.832547\n",
       "though         -2.802692\n",
       "perfect        -2.787033\n",
       "awesome        -2.723490\n",
       "next           -2.723467\n",
       "second         -2.653393\n",
       "unusual        -2.644315\n",
       "love           -2.643522\n",
       "easier         -2.522630\n",
       "Name: 1, dtype: float64"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model[\"1\"].sort_values(ascending=True).head(20)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Lastly, for scores as 3, we will not find words that indicate extreme (good or bad) reviews. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "wasted       -2.384733\n",
       "recommend    -2.159904\n",
       "power        -2.057191\n",
       "right        -2.018770\n",
       "describe     -2.014200\n",
       "barely       -2.010569\n",
       "wait         -1.998578\n",
       "thank        -1.941785\n",
       "daily        -1.924913\n",
       "shame        -1.917508\n",
       "save         -1.889566\n",
       "painful      -1.881169\n",
       "scientific   -1.848593\n",
       "drivel       -1.843792\n",
       "study        -1.828461\n",
       "loved        -1.824597\n",
       "extremely    -1.824438\n",
       "grader       -1.786883\n",
       "pull         -1.780412\n",
       "pointless    -1.745263\n",
       "Name: 3, dtype: float64"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model[\"3\"].sort_values(ascending=True).head(20)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Caveats\n",
    "\n",
    "This is a very simple approach to understand this exercise, so we can improve it a lot. We can reduce the dimensionality of the features, test different algorithms and tune the ML parameters. But in this notebook I show this simple solution to learn how it works."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.8.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}