Movie Review Sentiment Analyzer (NLP Project)

import tensorflow as tf
from tensorflow import keras
import numpy as np
import random

def build_nlp_model():
  """Loads text data, preprocesses it, and trains a Sentiment Analysis AI."""
  print("📥 Downloading IMDB Movie Reviews dataset...")
  
  # Keep only the top 10,000 most frequently occurring words in the training data
  vocab_size = 10000
  
  # Load the data. It comes pre-tokenized (words are already numbers)
  (x_train, y_train), (x_test, y_test) = keras.datasets.imdb.load_data(num_words=vocab_size)

  # 1. Preprocessing: Pad sequences
  # Some reviews are 50 words, some are 500. We force them all to be 250 words.
  print("⚙️ Padding sequences to uniform lengths...")
  max_length = 250
  x_train = keras.preprocessing.sequence.pad_sequences(x_train, maxlen=max_length, padding='post')
  x_test = keras.preprocessing.sequence.pad_sequences(x_test, maxlen=max_length, padding='post')

  # 2. Build the Keras Model
  print("🧠 Building the Keras NLP Neural Network...")
  model = keras.Sequential([
      # Embedding layer learns the 'meaning' of words based on their context
      keras.layers.Embedding(input_dim=vocab_size, output_dim=16, input_length=max_length),
      
      # Shrinks the 2D data down to 1D, making it easier for the next layer to process
      keras.layers.GlobalAveragePooling1D(),
      
      # A standard hidden layer to find patterns
      keras.layers.Dense(16, activation='relu'),
      
      # Output layer: 1 neuron. 
      # Sigmoid function outputs a probability between 0.0 (Negative) and 1.0 (Positive)
      keras.layers.Dense(1, activation='sigmoid')
  ])

  # 3. Compile the Model
  # Binary crossentropy is the standard loss function for Yes/No, True/False outputs
  model.compile(optimizer='adam', 
                loss='binary_crossentropy', 
                metrics=['accuracy'])

  # 4. Train the Model
  print("
🏋️‍♂️ Training the AI on 25,000 reviews (This takes a moment)...")
  # We use a validation split to ensure the AI isn't just memorizing the answers
  history = model.fit(x_train, y_train, epochs=3, batch_size=512, validation_split=0.2, verbose=1)

  print("
📊 Evaluating on unseen test data...")
  test_loss, test_acc = model.evaluate(x_test, y_test, verbose=0)
  print(f"🔹 Final AI Accuracy on brand new reviews: {test_acc * 100:.2f}%")

  return model, x_test, y_test

def decode_and_test(model, x_test, y_test):
  """Picks a random test review, decodes it to English, and prints the AI's prediction."""
  print("
" + "="*50)
  print("🎬 LIVE AI SENTIMENT TEST")
  print("="*50)

  # 1. Get the dictionary that maps numbers back to English words
  word_index = keras.datasets.imdb.get_word_index()
  
  # The IMDB dataset offsets word indices by 3 to reserve space for special characters
  reverse_word_index = {value + 3: key for key, value in word_index.items()}
  reverse_word_index[0] = "<PAD>"  # Padding
  reverse_word_index[1] = "<START>"
  reverse_word_index[2] = "<UNKNOWN>"

  def decode_review(encoded_text):
      return ' '.join([reverse_word_index.get(i, '?') for i in encoded_text])

  # 2. Pick a random review from the test set
  idx = random.randint(0, len(x_test) - 1)
  raw_review = x_test[idx]
  actual_sentiment = "Positive" if y_test[idx] == 1 else "Negative"

  # 3. Print the readable text
  english_review = decode_review(raw_review)
  
  # Clean up the padding tags so it is easier to read
  english_review = english_review.replace("<PAD> ", "").replace("<START> ", "")
  print(f"
📖 RANDOM REVIEW TEXT:
"{english_review[:400]}... [truncated]"")

  # 4. Ask the AI to predict the sentiment
  # We must reshape the data slightly for a single prediction
  prediction_data = np.expand_dims(raw_review, axis=0)
  ai_confidence = model.predict(prediction_data, verbose=0)[0][0]

  # If the score is > 0.5, it leans positive. If < 0.5, it leans negative.
  ai_prediction = "Positive" if ai_confidence >= 0.5 else "Negative"

  print("
" + "-"*50)
  print(f"🎯 ACTUAL SENTIMENT: {actual_sentiment}")
  print(f"🤖 AI PREDICTION:    {ai_prediction} (Score: {ai_confidence:.4f})")
  
  if actual_sentiment == ai_prediction:
      print("✅ The AI understood the emotion correctly!")
  else:
      print("❌ The AI was confused by the wording.")
  print("="*50 + "
")

if __name__ == "__main__":
  trained_model, test_reviews, test_labels = build_nlp_model()
  decode_and_test(trained_model, test_reviews, test_labels)