{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/018e07c13ff74f4f92fb6efe7ff64693\" frameborder=\"0\" width=\"1920\" height=\"1440\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1440,"width":1920,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1440,"thumbnail_width":1920,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/018e07c13ff74f4f92fb6efe7ff64693-1ecf2278cbc6d700.gif","duration":580.289,"title":"Advanced Retrieval Methods for RAG","description":"This video reviews advanced retrieval methods for RAG, comparing BM25, dense retrieval, and hybrid ranking approaches. It explains BM25’s strength for keyword based queries using inverse document frequency, showing cases where BM25 outperformed dense retrieval on the first query but dense retrieval did better on the second, while both struggled on the third. It covers improvements like child-parent retrieval for richer context and hybrid fusion with Reciprocal Ranked Fusion that scores documents by rank rather than similarity scale, as well as Cohere ranking that reranks pages using a model for contextual relevance. It concludes with multi-query retrieval, where an LLM generates alternate queries and repeatedly retrieved chunks are considered more relevant, noting it is the heaviest method due to extra model calls."}