Familia

Why we included this project

Baidu built Familia so teams can apply topic modeling to production text without first spending months training models. It ships LDA, SentenceLDA, and Topical Word Embedding implementations trained on large industrial corpora, plus tools for document topic inference and semantic matching, so you can pull topic vectors, extract keywords, compare document similarity, or match a search query against a page. Pretrained models cover news, novels, and web pages, and you can train your own on a custom corpus if your data is different. For text classification, clustering, and recommendation, it offers interpretable topical structure rather than a black-box embedding. One caveat: the repo has been quiet since 2021, so treat it as mature and stable rather than actively developed.