lgalke/vec4ir

Implements multiple embedding-based retrieval models (Word Centroid Similarity, IDF-reweighted variants) integrated with gensim for training Skip-gram and GloVe embeddings. The framework provides a modular pipeline for matching and similarity scoring with built-in evaluation metrics, designed for extensibility through sklearn-inspired APIs to benchmark custom retrieval models against standard IR benchmarks.

226 stars. No commits in the last 6 months.