{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "dfb76af9",
   "metadata": {},
   "source": [
    "# Sentiment Analysis Using Classical ML\n",
    "\n",
    "**Course:** MIAI507 – Natural Language Processing  \n",
    "**Student:** Hadi Hijazi\n",
    "\n",
    "In this notebook I classify IMDB movie reviews as positive or negative. I keep the classifier the same in every experiment (Logistic Regression) and only change how the text is turned into numbers:\n",
    "\n",
    "1. word counts (CountVectorizer)\n",
    "2. TF-IDF weights\n",
    "3. pretrained word embeddings (GloVe)\n",
    "4. word embeddings I train myself on the reviews (word2vec)\n",
    "\n",
    "This way, any difference in accuracy comes from the features, not from the model."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1f81caeb",
   "metadata": {},
   "source": [
    "## Setup\n",
    "\n",
    "All the imports are here. I also wrote a small helper, `evaluate`, because Tasks 3 to 6 all repeat the same steps: train the pipeline, test it, and print the accuracy and the classification report. It also saves each accuracy in `results`, which I use for the chart in Task 7."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "72e815a1",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-11-08T15:57:10.995921Z",
     "iopub.status.busy": "2026-11-08T15:57:10.995684Z",
     "iopub.status.idle": "2026-11-08T15:57:16.194870Z",
     "shell.execute_reply": "2026-11-08T15:57:16.194244Z"
    }
   },
   "outputs": [],
   "source": [
    "import io\n",
    "import os\n",
    "import re\n",
    "from contextlib import redirect_stderr\n",
    "\n",
    "import gensim.downloader as api\n",
    "import kagglehub\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from gensim.models import Word2Vec\n",
    "from sklearn.base import BaseEstimator, TransformerMixin\n",
    "from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.metrics import accuracy_score, classification_report\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.pipeline import make_pipeline\n",
    "\n",
    "RANDOM_STATE = 42\n",
    "results = {}\n",
    "\n",
    "\n",
    "def evaluate(name, model):\n",
    "    \"\"\"Train the model, print its test accuracy and classification report, and store the accuracy.\"\"\"\n",
    "    model.fit(X_train, y_train)\n",
    "    y_pred = model.predict(X_test)\n",
    "    accuracy = accuracy_score(y_test, y_pred)\n",
    "    results[name] = accuracy\n",
    "    print(f\"{name}\\nAccuracy: {accuracy:.4f}\\n\")\n",
    "    print(classification_report(y_test, y_pred, target_names=[\"negative\", \"positive\"], digits=4))\n",
    "    return model"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "826687e0",
   "metadata": {},
   "source": [
    "## Task 1 – Load the IMDB Movie Review Dataset\n",
    "\n",
    "The dataset is the [IMDB Dataset of 50K Movie Reviews](https://www.kaggle.com/datasets/lakshmi25npathi/imdb-dataset-of-50k-movie-reviews) on Kaggle. Instead of downloading the CSV by hand, I use `kagglehub`: it downloads the file once and caches it, so the notebook runs the same on any machine. Then I load it with pandas."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "31f73095",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-11-08T15:57:16.196450Z",
     "iopub.status.busy": "2026-11-08T15:57:16.196301Z",
     "iopub.status.idle": "2026-11-08T15:57:16.955325Z",
     "shell.execute_reply": "2026-11-08T15:57:16.954783Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape: (50000, 2)\n",
      "sentiment\n",
      "positive    25000\n",
      "negative    25000 \n",
      "\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>review</th>\n",
       "      <th>sentiment</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>One of the other reviewers has mentioned that ...</td>\n",
       "      <td>positive</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>A wonderful little production. &lt;br /&gt;&lt;br /&gt;The...</td>\n",
       "      <td>positive</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>I thought this was a wonderful way to spend ti...</td>\n",
       "      <td>positive</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Basically there's a family where a little boy ...</td>\n",
       "      <td>negative</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Petter Mattei's \"Love in the Time of Money\" is...</td>\n",
       "      <td>positive</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                              review sentiment\n",
       "0  One of the other reviewers has mentioned that ...  positive\n",
       "1  A wonderful little production. <br /><br />The...  positive\n",
       "2  I thought this was a wonderful way to spend ti...  positive\n",
       "3  Basically there's a family where a little boy ...  negative\n",
       "4  Petter Mattei's \"Love in the Time of Money\" is...  positive"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset_dir = kagglehub.dataset_download(\"lakshmi25npathi/imdb-dataset-of-50k-movie-reviews\")\n",
    "df = pd.read_csv(os.path.join(dataset_dir, \"IMDB Dataset.csv\"))\n",
    "\n",
    "print(\"Shape:\", df.shape)\n",
    "print(df[\"sentiment\"].value_counts().to_string(), \"\\n\")\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "535788f0",
   "metadata": {},
   "source": [
    "### A first look at the reviews\n",
    "\n",
    "The head cuts every review after about 50 characters, which is not enough to judge what the text looks like. Before cleaning anything, I want to read one review in full."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "abe64bd7",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-11-08T15:57:16.956703Z",
     "iopub.status.busy": "2026-11-08T15:57:16.956607Z",
     "iopub.status.idle": "2026-11-08T15:57:16.961653Z",
     "shell.execute_reply": "2026-11-08T15:57:16.961179Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "A wonderful little production. <br /><br />The filming technique is very unassuming- very old-time-BBC fashion and gives a comforting, and sometimes discomforting, sense of realism to the entire piece. <br /><br />The actors are extremely well chosen- Michael Sheen not only \"has got all the polari\" but he has all the voices down pat too! You can truly see the seamless editing guided by the references to Williams' diary entries, not only is it well worth the watching but it is a terrificly written and performed piece. A masterful production about one of the great master's of comedy and his life. <br /><br />The realism really comes home with the little things: the fantasy of the guard which, rather than use the traditional 'dream' techniques remains solid then disappears. It plays on our knowledge and our senses, particularly with the scenes concerning Orton and Halliwell and the sets (particularly of their flat with Halliwell's murals decorating every surface) are terribly well done.\n"
     ]
    }
   ],
   "source": [
    "print(df.loc[1, \"review\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b18617ad",
   "metadata": {},
   "source": [
    "The review itself is normal English, but there are `<br /><br />` tags in the middle of it. The reviews were collected from a web page and kept its HTML line breaks. If I leave them, `CountVectorizer` will count *br* as a word in every review that has them. How many reviews are affected?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "d5f6a40f",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-11-08T15:57:16.963054Z",
     "iopub.status.busy": "2026-11-08T15:57:16.962966Z",
     "iopub.status.idle": "2026-11-08T15:57:17.000845Z",
     "shell.execute_reply": "2026-11-08T15:57:17.000431Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "29,200 of 50,000 reviews contain <br /> tags (58%)\n"
     ]
    }
   ],
   "source": [
    "has_br = df[\"review\"].str.contains(\"<br\", regex=False)\n",
    "print(f\"{has_br.sum():,} of {len(df):,} reviews contain <br /> tags ({has_br.mean():.0%})\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a0a3f237",
   "metadata": {},
   "source": [
    "More than half of them, so the tags have to go.\n",
    "\n",
    "With 50,000 reviews collected from a website, I also want to check whether some were collected twice."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "2cbd0b12",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-11-08T15:57:17.002267Z",
     "iopub.status.busy": "2026-11-08T15:57:17.002183Z",
     "iopub.status.idle": "2026-11-08T15:57:17.040747Z",
     "shell.execute_reply": "2026-11-08T15:57:17.040386Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "418 rows are copies of a review that already appears earlier in the dataset\n",
      "\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>review</th>\n",
       "      <th>sentiment</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>34058</th>\n",
       "      <td>\"Go Fish\" garnered Rose Troche rightly or wron...</td>\n",
       "      <td>negative</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>47467</th>\n",
       "      <td>\"Go Fish\" garnered Rose Troche rightly or wron...</td>\n",
       "      <td>negative</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>29956</th>\n",
       "      <td>\"Three\" is a seriously dumb shipwreck movie. M...</td>\n",
       "      <td>negative</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>31488</th>\n",
       "      <td>\"Three\" is a seriously dumb shipwreck movie. M...</td>\n",
       "      <td>negative</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>47527</th>\n",
       "      <td>\"Witchery\" might just be the most incoherent a...</td>\n",
       "      <td>negative</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2976</th>\n",
       "      <td>\"Witchery\" might just be the most incoherent a...</td>\n",
       "      <td>negative</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                  review sentiment\n",
       "34058  \"Go Fish\" garnered Rose Troche rightly or wron...  negative\n",
       "47467  \"Go Fish\" garnered Rose Troche rightly or wron...  negative\n",
       "29956  \"Three\" is a seriously dumb shipwreck movie. M...  negative\n",
       "31488  \"Three\" is a seriously dumb shipwreck movie. M...  negative\n",
       "47527  \"Witchery\" might just be the most incoherent a...  negative\n",
       "2976   \"Witchery\" might just be the most incoherent a...  negative"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "n_copies = df.duplicated(subset=\"review\").sum()\n",
    "print(f\"{n_copies} rows are copies of a review that already appears earlier in the dataset\\n\")\n",
    "df[df.duplicated(subset=\"review\", keep=False)].sort_values(\"review\").head(6)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f43cb75a",
   "metadata": {},
   "source": [
    "There are 418 copies, and the table shows the same text repeated with the same label. If one copy ends up in the training set and another in the test set, the model is tested on a review it has already seen, and the accuracy looks better than it really is. I remove them in Task 2.\n",
    "\n",
    "Last, I look at how long the reviews are. This matters for Task 5, where one of the approaches only keeps the first 200 words of each review."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "2e82b351",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-11-08T15:57:17.042192Z",
     "iopub.status.busy": "2026-11-08T15:57:17.042103Z",
     "iopub.status.idle": "2026-11-08T15:57:17.916495Z",
     "shell.execute_reply": "2026-11-08T15:57:17.915909Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "count    50000.0\n",
      "mean       231.2\n",
      "std        171.3\n",
      "min          4.0\n",
      "25%        126.0\n",
      "50%        173.0\n",
      "75%        280.0\n",
      "max       2470.0\n",
      "\n",
      "Reviews longer than 200 words: 41%\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 900x400 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "lengths = df[\"review\"].str.split().str.len()\n",
    "print(lengths.describe().round(1).to_string())\n",
    "print(f\"\\nReviews longer than 200 words: {(lengths > 200).mean():.0%}\")\n",
    "\n",
    "plt.figure(figsize=(9, 4))\n",
    "plt.hist(lengths, bins=100, range=(0, 1000), color=\"steelblue\")\n",
    "plt.axvline(200, color=\"darkorange\", linestyle=\"--\", label=\"200 words\")\n",
    "plt.xlabel(\"Words per review\")\n",
    "plt.ylabel(\"Number of reviews\")\n",
    "plt.title(\"Review length (reviews above 1,000 words not shown)\")\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1a972102",
   "metadata": {},
   "source": [
    "The median review has 173 words, but the distribution has a long tail: about 40% of the reviews are longer than 200 words, and the longest has almost 2,500. I come back to this in Task 5.\n",
    "\n",
    "There is also a small bump right before 1,000 words, and very few reviews after it. My guess is that IMDB limited reviews to about 1,000 words when most of these were written, so people with a lot to say wrote right up to the limit."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "98e05fa4",
   "metadata": {},
   "source": [
    "## Task 2 – Split the Dataset into Training and Testing\n",
    "\n",
    "Based on what I saw above, I clean two things before splitting:\n",
    "\n",
    "- I drop the 418 duplicate reviews.\n",
    "- I replace the `<br />` tags with spaces.\n",
    "\n",
    "Then I separate the features (the review text) from the targets (the sentiment, converted to 1 for positive and 0 for negative) and split them 60% / 40%. I use `stratify=y` so both sets keep the 50/50 balance between positive and negative reviews, and a fixed `random_state` so I get the same split every time I run the notebook."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "785da53e",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-11-08T15:57:17.918101Z",
     "iopub.status.busy": "2026-11-08T15:57:17.918006Z",
     "iopub.status.idle": "2026-11-08T15:57:18.040844Z",
     "shell.execute_reply": "2026-11-08T15:57:18.040360Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Removed 418 duplicate reviews, 49,582 remain\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training examples: 29,749\n",
      "Testing examples:  19,833\n"
     ]
    }
   ],
   "source": [
    "n_before = len(df)\n",
    "df = df.drop_duplicates(subset=\"review\")\n",
    "print(f\"Removed {n_before - len(df)} duplicate reviews, {len(df):,} remain\\n\")\n",
    "\n",
    "X = df[\"review\"].str.replace(r\"<br\\s*/?>\", \" \", regex=True)\n",
    "y = (df[\"sentiment\"] == \"positive\").astype(int)\n",
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(\n",
    "    X, y, test_size=0.4, stratify=y, random_state=RANDOM_STATE\n",
    ")\n",
    "print(f\"Training examples: {len(X_train):,}\")\n",
    "print(f\"Testing examples:  {len(X_test):,}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c8c6edcb",
   "metadata": {},
   "source": [
    "## Task 3 – Count Vectorizer + Logistic Regression\n",
    "\n",
    "I start with the simplest representation. `CountVectorizer` builds a vocabulary from the training reviews and turns each review into a vector of word counts (a bag of words). Word order is lost; only the counts remain. Logistic Regression then learns one weight per word.\n",
    "\n",
    "I set `max_iter=1000` because with the default of 100 iterations, the solver stops before it converges on this many features."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "0d0c8c7b",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-11-08T15:57:18.042235Z",
     "iopub.status.busy": "2026-11-08T15:57:18.042139Z",
     "iopub.status.idle": "2026-11-08T15:57:27.550172Z",
     "shell.execute_reply": "2026-11-08T15:57:27.549541Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Count + LR\n",
      "Accuracy: 0.8857\n",
      "\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "    negative     0.8942    0.8741    0.8840      9879\n",
      "    positive     0.8778    0.8973    0.8874      9954\n",
      "\n",
      "    accuracy                         0.8857     19833\n",
      "   macro avg     0.8860    0.8857    0.8857     19833\n",
      "weighted avg     0.8859    0.8857    0.8857     19833\n",
      "\n"
     ]
    }
   ],
   "source": [
    "count_model = evaluate(\n",
    "    \"Count + LR\",\n",
    "    make_pipeline(CountVectorizer(), LogisticRegression(max_iter=1000)),\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7fc2b967",
   "metadata": {},
   "source": [
    "## Task 4 – TF-IDF Vectorizer + Logistic Regression\n",
    "\n",
    "Raw counts treat every word the same. But a word like *movie* appears in almost every review, so it says very little about the sentiment. `TfidfTransformer` corrects this: it gives less weight to words that appear in many reviews and more weight to rarer ones. It also scales each review vector to length 1, so a long review doesn't get bigger numbers just because it has more words."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "43c15814",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-11-08T15:57:27.551491Z",
     "iopub.status.busy": "2026-11-08T15:57:27.551417Z",
     "iopub.status.idle": "2026-11-08T15:57:31.014761Z",
     "shell.execute_reply": "2026-11-08T15:57:31.014158Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "TF-IDF + LR\n",
      "Accuracy: 0.8926\n",
      "\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "    negative     0.9059    0.8752    0.8903      9879\n",
      "    positive     0.8802    0.9098    0.8947      9954\n",
      "\n",
      "    accuracy                         0.8926     19833\n",
      "   macro avg     0.8930    0.8925    0.8925     19833\n",
      "weighted avg     0.8930    0.8926    0.8925     19833\n",
      "\n"
     ]
    }
   ],
   "source": [
    "tfidf_model = evaluate(\n",
    "    \"TF-IDF + LR\",\n",
    "    make_pipeline(CountVectorizer(), TfidfTransformer(), LogisticRegression(max_iter=1000)),\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "21ee6fe8",
   "metadata": {},
   "source": [
    "## Task 5 – Pretrained Word Embeddings + Logistic Regression\n",
    "\n",
    "Here the idea changes. Instead of one feature per word, each word becomes a dense vector that represents its meaning, learned beforehand on a large amount of text.\n",
    "\n",
    "**Pretrained model.** I use GloVe from gensim (`glove-wiki-gigaword-100`: 400,000 words, trained on Wikipedia and news articles). I chose the 100-dimensional version on purpose: with concatenation, each review becomes 200 × 100 = 20,000 numbers. With the 300-dimensional version it would be 60,000 numbers per review, about 7 GB for the training set alone, which is too much for my laptop.\n",
    "\n",
    "**Embedding vectorizer.** A scikit-learn pipeline only accepts steps that have `fit` and `transform` methods, so I wrote `EmbeddingVectorizer`, a small class that inherits from `BaseEstimator` and `TransformerMixin`. `fit` has nothing to learn, since the embeddings are already trained. `transform` lowercases each review, splits it into words, skips the words GloVe doesn't know, and turns the rest into one vector. I try the two approaches from the assignment and keep the better one:\n",
    "\n",
    "- **Mean:** average the vectors of all the words, giving 100 numbers per review.\n",
    "- **Concatenation:** place the vectors of the first 200 words one after another, padding with zeros when the review is shorter, giving 20,000 numbers per review."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "7df90272",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-11-08T15:57:31.016193Z",
     "iopub.status.busy": "2026-11-08T15:57:31.016094Z",
     "iopub.status.idle": "2026-11-08T15:59:15.425162Z",
     "shell.execute_reply": "2026-11-08T15:59:15.424223Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GloVe: 400,000 words, 100 dimensions\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GloVe mean + LR\n",
      "Accuracy: 0.7985\n",
      "\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "    negative     0.7964    0.8000    0.7982      9879\n",
      "    positive     0.8006    0.7970    0.7988      9954\n",
      "\n",
      "    accuracy                         0.7985     19833\n",
      "   macro avg     0.7985    0.7985    0.7985     19833\n",
      "weighted avg     0.7985    0.7985    0.7985     19833\n",
      "\n",
      "------------------------------------------------------------\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GloVe concat + LR\n",
      "Accuracy: 0.6836\n",
      "\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "    negative     0.6834    0.6795    0.6815      9879\n",
      "    positive     0.6837    0.6876    0.6856      9954\n",
      "\n",
      "    accuracy                         0.6836     19833\n",
      "   macro avg     0.6836    0.6835    0.6835     19833\n",
      "weighted avg     0.6836    0.6836    0.6836     19833\n",
      "\n",
      "------------------------------------------------------------\n",
      "Best approach with GloVe: mean\n"
     ]
    }
   ],
   "source": [
    "TOKEN_PATTERN = re.compile(r\"[a-z]+\")\n",
    "\n",
    "\n",
    "def tokenize(text):\n",
    "    return TOKEN_PATTERN.findall(text.lower())\n",
    "\n",
    "\n",
    "class EmbeddingVectorizer(BaseEstimator, TransformerMixin):\n",
    "    \"\"\"Turn each review into a vector using word embeddings (gensim KeyedVectors).\"\"\"\n",
    "\n",
    "    def __init__(self, embeddings, method=\"mean\", max_words=200):\n",
    "        self.embeddings = embeddings\n",
    "        self.method = method\n",
    "        self.max_words = max_words\n",
    "\n",
    "    def fit(self, X, y=None):\n",
    "        return self\n",
    "\n",
    "    def transform(self, X):\n",
    "        dim = self.embeddings.vector_size\n",
    "        n_features = dim if self.method == \"mean\" else dim * self.max_words\n",
    "        vectors = np.zeros((len(X), n_features), dtype=np.float32)\n",
    "        for i, text in enumerate(X):\n",
    "            words = [w for w in tokenize(text) if w in self.embeddings]\n",
    "            if not words:\n",
    "                continue\n",
    "            if self.method == \"mean\":\n",
    "                vectors[i] = self.embeddings[words].mean(axis=0)\n",
    "            else:\n",
    "                words = words[: self.max_words]\n",
    "                vectors[i, : len(words) * dim] = self.embeddings[words].ravel()\n",
    "        return vectors\n",
    "\n",
    "\n",
    "glove = api.load(\"glove-wiki-gigaword-100\")\n",
    "print(f\"GloVe: {len(glove):,} words, {glove.vector_size} dimensions\\n\")\n",
    "\n",
    "for method in [\"mean\", \"concat\"]:\n",
    "    evaluate(\n",
    "        f\"GloVe {method} + LR\",\n",
    "        make_pipeline(EmbeddingVectorizer(glove, method), LogisticRegression(max_iter=1000)),\n",
    "    )\n",
    "    print(\"-\" * 60)\n",
    "\n",
    "best = max([\"mean\", \"concat\"], key=lambda m: results[f\"GloVe {m} + LR\"])\n",
    "print(f\"Best approach with GloVe: {best}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "63021c5a",
   "metadata": {},
   "source": [
    "## Task 6 – Own Word2Vec Embeddings + Logistic Regression\n",
    "\n",
    "Now I train my own word2vec model (CBOW, gensim's default) on the IMDB reviews. I train it on the **training set only**, so the test reviews stay completely unseen, like in the other tasks.\n",
    "\n",
    "I keep 100 dimensions so the comparison with GloVe is fair, use a window of 5 words around each word, ignore words that appear fewer than 5 times, and go over the data 10 times.\n",
    "\n",
    "Before using it, I check that the model learned something meaningful by printing the closest words to *excellent* and *boring*. Then I reuse the same `EmbeddingVectorizer` and both approaches from Task 5."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "af1cea8c",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-11-08T15:59:15.429344Z",
     "iopub.status.busy": "2026-11-08T15:59:15.429133Z",
     "iopub.status.idle": "2026-11-08T16:00:08.617738Z",
     "shell.execute_reply": "2026-11-08T16:00:08.617027Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Own word2vec: 31,467 words, 100 dimensions\n",
      "  most similar to 'excellent': outstanding, exceptional, superb, terrific, amazing, incredible\n",
      "  most similar to 'boring': dull, pointless, tedious, predictable, confusing, unoriginal\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Word2Vec mean + LR\n",
      "Accuracy: 0.8525\n",
      "\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "    negative     0.8573    0.8445    0.8508      9879\n",
      "    positive     0.8479    0.8605    0.8542      9954\n",
      "\n",
      "    accuracy                         0.8525     19833\n",
      "   macro avg     0.8526    0.8525    0.8525     19833\n",
      "weighted avg     0.8526    0.8525    0.8525     19833\n",
      "\n",
      "------------------------------------------------------------\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Word2Vec concat + LR\n",
      "Accuracy: 0.7534\n",
      "\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "    negative     0.7536    0.7502    0.7519      9879\n",
      "    positive     0.7532    0.7566    0.7549      9954\n",
      "\n",
      "    accuracy                         0.7534     19833\n",
      "   macro avg     0.7534    0.7534    0.7534     19833\n",
      "weighted avg     0.7534    0.7534    0.7534     19833\n",
      "\n",
      "------------------------------------------------------------\n",
      "Best approach with own word2vec: mean\n"
     ]
    }
   ],
   "source": [
    "train_sentences = [tokenize(review) for review in X_train]\n",
    "\n",
    "# gensim's compiled training code sometimes prints a harmless, empty\n",
    "# \"Exception ignored in: ...our_dot_float\" message (no real exception occurs),\n",
    "# so its stderr output is hidden during training only.\n",
    "with redirect_stderr(io.StringIO()):\n",
    "    w2v = Word2Vec(\n",
    "        train_sentences,\n",
    "        vector_size=100,\n",
    "        window=5,\n",
    "        min_count=5,\n",
    "        epochs=10,\n",
    "        workers=4,\n",
    "        seed=RANDOM_STATE,\n",
    "    )\n",
    "own_embeddings = w2v.wv\n",
    "print(f\"Own word2vec: {len(own_embeddings):,} words, {own_embeddings.vector_size} dimensions\")\n",
    "for word in [\"excellent\", \"boring\"]:\n",
    "    similar = \", \".join(w for w, _ in own_embeddings.most_similar(word, topn=6))\n",
    "    print(f\"  most similar to '{word}': {similar}\")\n",
    "print()\n",
    "\n",
    "for method in [\"mean\", \"concat\"]:\n",
    "    evaluate(\n",
    "        f\"Word2Vec {method} + LR\",\n",
    "        make_pipeline(EmbeddingVectorizer(own_embeddings, method), LogisticRegression(max_iter=1000)),\n",
    "    )\n",
    "    print(\"-\" * 60)\n",
    "\n",
    "best = max([\"mean\", \"concat\"], key=lambda m: results[f\"Word2Vec {m} + LR\"])\n",
    "print(f\"Best approach with own word2vec: {best}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f09ffa60",
   "metadata": {},
   "source": [
    "## Task 7 – Plot and Interpret the Results\n",
    "\n",
    "One bar per model, colored by family: bag of words in blue, GloVe in orange, my own word2vec in green."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "b3c97195",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-11-08T16:00:08.621395Z",
     "iopub.status.busy": "2026-11-08T16:00:08.621248Z",
     "iopub.status.idle": "2026-11-08T16:00:08.733258Z",
     "shell.execute_reply": "2026-11-08T16:00:08.732718Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1100x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "names = list(results)\n",
    "accuracies = [results[n] for n in names]\n",
    "colors = [\"steelblue\", \"steelblue\", \"darkorange\", \"darkorange\", \"seagreen\", \"seagreen\"]\n",
    "\n",
    "fig, ax = plt.subplots(figsize=(11, 6))\n",
    "bars = ax.bar(names, accuracies, color=colors)\n",
    "ax.bar_label(bars, labels=[f\"{a:.2%}\" for a in accuracies], padding=3)\n",
    "ax.set_ylim(0.4, 1.0)\n",
    "ax.axhline(0.5, color=\"gray\", linestyle=\"--\", linewidth=1, label=\"Random guess (50%)\")\n",
    "ax.legend(loc=\"upper right\")\n",
    "ax.set_ylabel(\"Test accuracy\")\n",
    "ax.set_title(\"Logistic Regression accuracy by text representation\")\n",
    "ax.spines[[\"top\", \"right\"]].set_visible(False)\n",
    "plt.xticks(rotation=15)\n",
    "plt.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9e88588a",
   "metadata": {},
   "source": [
    "### What the results show\n",
    "\n",
    "*The y-axis starts at 40% so the differences are easier to see. The dashed line is a random guess (50%, since the classes are balanced). My word2vec results can move by about 0.1% between runs, because gensim trains with several threads and that is not exactly reproducible.*\n",
    "\n",
    "**The simplest models win.** TF-IDF is the best (89.3%), just ahead of plain counts (88.6%). I think the reason is the way people express sentiment in reviews: through specific words like *excellent*, *worst*, *waste* or *masterpiece*. With a bag of words, each of the ~83,000 words in the vocabulary gets its own weight, so the model can pick these words up directly. TF-IDF adds 0.7 points on top, by lowering the weight of words that are everywhere (*movie*, *film*) and by putting long and short reviews on the same scale.\n",
    "\n",
    "**Averaging embeddings loses information.** With averaging, a review of about 230 words becomes just 100 numbers. The few words that carry the opinion get mixed with all the neutral words about the plot and the actors, and their signal is diluted. There is a second problem I didn't expect: embeddings put words close together when they appear in similar contexts, and opposite opinions appear in exactly the same contexts. In GloVe, the 2nd closest word to *bad* is *good* (similarity 0.77). So a positive and a negative review can end up with similar averages.\n",
    "\n",
    "**My own embeddings beat GloVe by more than 5 points (about 85% vs 79.9%).** GloVe learned its vectors from Wikipedia and news, not from people talking about movies. My word2vec model learned from the reviews themselves, and it shows: its closest words to *boring* include *dull*, *pointless*, *tedious* and *predictable*, exactly how reviewers write. It knows only 31,000 words against GloVe's 400,000, but they are the right words for this task.\n",
    "\n",
    "**Concatenation is the worst approach for both (68.4% with GloVe, about 75% with my word2vec).** The problem is position. *Excellent* as the 3rd word and *excellent* as the 50th word land on completely different features, so the model has to learn the effect of each word separately for each of the 200 positions: 20,000 features for only about 30,000 training reviews. On top of that, as I saw when exploring the data, about 40% of the reviews are longer than 200 words, so their ending is cut off, and that is often where the reviewer gives the final verdict. To really use word order, a sequence model (RNN, LSTM or transformer) would be needed.\n",
    "\n",
    "**Conclusion.** For a linear classifier on this dataset, TF-IDF is the best choice: the most accurate, and also the simplest and fastest to train. Word embeddings only start to compete when they are trained on the same kind of text, and even then, averaging them throws away too much to beat a plain bag of words."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": ".venv",
   "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": "3.12.14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}
