A sub-intent is a distinct information need implied by a broader query. Treating one query as one guess at user intent can miss documents that answer only part of the question. section 8.0 section 8.7
RAG-Fusion uses sub-intents to expand retrieval: an LLM decomposes the original question into perspective-shifted queries, which run in parallel before their result lists are merged with Reciprocal Rank Fusion. section 8.0 section 8.2
Useful sub-intents shift perspective rather than merely rephrase the same request:
- Build quality: durable laptops for students. section 8.2
- Portability: lightweight laptops for carrying to class. section 8.2
- Price: best budget laptops for college. section 8.2
- Battery life: laptops for studying with long battery life. section 8.2
From broad query to retrieval coverage
For “What is the best laptop for college?”, simple rephrasings tend to retrieve the same documents. Queries focused on durability, portability, price, and battery life reach different document sets, widening coverage while remaining tied to the original need. section 8.2
You are a search assistant. The user has asked: "{question}". Generate {N} alternative search queries that target different facets, sub-intents, or interpretations of this question. Each query should explore a distinct angle so that together they cover the full intent. Return one query per line, no numbering.
Use the prompt template to instruct an LLM to generate one search query per distinct facet, sub-intent, or interpretation. section 8.2
