LittletreeZou/Question-Pairs-Matching

Implements a semantic similarity matching pipeline for customer service questions using ensemble methods combining word embeddings, textual features, and deep learning models (CNNs/RNNs). The solution preprocesses text with tokenization and synonym expansion, then trains multiple classifiers whose predictions are aggregated to classify question pairs as duplicates or distinct. Targets the MoJing Cup competition dataset with detailed ablation studies documenting feature engineering and model selection choices available in the accompanying technical write-up.

148 stars. No commits in the last 6 months.