Enriched indexing of content chunks
Key Takeaway
Enriched indexing, turned on with the Enrich Content Chunks toggle, uses an LLM to add a metadata chunk and a question chunk for every chunk in a vector or hybrid search index. Search can then match a user's question to a passage that is worded very differently, and the results still return the original chunk text.
Enriched indexing is an option on a Data 360 vector or hybrid search index. It uses an LLM to generate extra content for each chunk of your unstructured data, such as keywords, a summary and the questions the chunk can answer. With this extra content, search can match a user's question to the passage that holds the answer, even when that passage is worded very differently from the question.
How enriched indexing extends a search index
When Data 360 builds a vector or hybrid search index, it breaks the source content into semantically related chunks. Then it creates searchable vectors for those chunks.
Enriched indexing adds one more step: it uses an LLM to extract metadata from each chunk. You turn on enriched indexing with the "Enrich Content Chunks" toggle when you create the index.