← All questions
Retrieval & EmbeddingsEasy· Doc2Query· Vocabulary Mismatch· BM25

Why would generating queries from a document help retri...

The primary challenge in traditional retrieval is vocabulary mismatch, where a document is relevant but does not contain the specific keywords a user enters. BM25-style scoring contributes zero relevance for absent terms, regardless of how semantically related the document is. By predicting and appending potential queries to the document at index time, you change the indexed content to include these missing terms, effectively bridging the lexical gap between user intent and document language.

Code playground (Python)