https://weaviate.io/blog/hybrid-search-explained
搜索可以用 hybrid: 一個向量搜尋 + 一個BM25
兩種融合方式
- Relative Score Fusion: an alpha term between 0 and 1 weighing dense vector similarity w/ sparse vector similarity
- Reciprocal Rank Fusion: add 1/rank in the ranked list for both sparse and dense retrieval.
跟 [[Ensemble Retriever]] 其實一樣
只是 Hybrid Search 通常指直接用 Vector Store 提供的 Hybrid Search 功能
Ensemble Retriever 是自幹
e.g.
https://python.langchain.com/docs/integrations/retrievers/weaviate-hybrid
介紹文章: https://towardsdatascience.com/improving-retrieval-performance-in-rag-pipelines-with-hybrid-search-c75203c2f2f5#f7da
llamaindex 的 Qdrant Hybrid Search 範例
https://twitter.com/jerryjliu0/status/1738583302481842339
## LlamaIndex:在RAG中通過Alpha調整增強混合檢索的檢索性能
https://www.llamaindex.ai/blog/llamaindex-enhancing-retrieval-performance-with-alpha-tuning-in-hybrid-search-in-rag-135d0c9b8a00
https://twitter.com/llama_index/status/1763252392639042024