0xku/information-retrieval

Comprehensive tutorial collection covering the full IR pipeline: from classical inverted index methods through modern dense retrieval architectures including bi-encoders, cross-encoders, and multilingual variants. Covers evaluation metrics (MRR, MAP, nDCG), dense representation learning from LSA to transformer finetuning, and unsupervised training approaches (TSDAE, SimCSE, GPL) that reduce labeled data requirements. Integrates BERT and Sentence-BERT frameworks with approximate nearest neighbor indexing techniques for scalable vector search across millions of documents.

174 stars. No commits in the last 6 months.