boudinfl/pke

Implements both unsupervised (statistical and graph-based) and supervised keyphrase extraction models with a modular pipeline where candidate selection, weighting, and ranking stages can be independently customized. Leverages spaCy for NLP preprocessing and includes pre-trained models on SemEval-2010, enabling immediate benchmarking against state-of-the-art approaches like TextRank, TopicRank, and MultipartiteRank.

1,590 stars. No commits in the last 6 months.