Semantic Similarity Detection Using Deep Learning and Feature Engineering

Built and evaluated multiple machine learning models to detect semantically duplicate questions using the Quora dataset. Compared traditional classifiers using hand-engineered similarity features with deep learning models including Siamese LSTM and Bilateral Multi-Perspective Matching (BiMPM). Explored GloVe and Word2Vec embeddings and developed a hybrid architecture combining CNN, LSTM, and engineered features. The best model—fine-tuned GloVe with LSTM+CNN and baseline features—achieved a log-loss of 0.21 on the Kaggle test set, significantly outperforming standard approaches. This project highlights the effectiveness of blending deep learning with traditional NLP features for semantic similarity tasks.

For details View Report