3 ms·
Did this in the summer via RAG. One thing we realised is that pure vector embeddings retrieval doesn't work so well for docs with acronyms (which let's face it
by monkeydust 3y ago
Did this in the summer via RAG. One thing we realised is that pure vector embeddings retrieval doesn't work so well for docs with acronyms (which let's face it all businesses have). Created a hybrid solution using embeddings and BM25 which is traditional ranking tool. This hybrid gave best results.
- gardnr 3y agoI was going to ask how you integrated BM25 but then I found this: https://docs.llamaindex.ai/en/stable/examples/retrievers/bm25_retriever.html https://docs.llamaindex.ai/en/stable/examples/retrievers/bm2...