jainarika/sentiment-lstm

Sentiment analysis with a Bidirectional LSTM

Binary sentiment classifier (positive / negative) for movie reviews, trained on the IMDB dataset.

Model

• Embedding (20,000-word vocabulary, masked padding) -> SpatialDropout -> 2 stacked Bidirectional LSTM layers -> Dense -> sigmoid
• Input: reviews cleaned, tokenised and padded to 400 tokens

Results

• Test accuracy: 66.80%

Files

• best_model.keras: trained Keras model
• word_index.pkl: vocabulary used for tokenising
• app.log: training log

Limitations

Trained only on movie reviews; may do poorly on other domains, sarcasm, or mixed-sentiment text.