query rewriting

**Query rewriting** is the **transformation of user queries into clearer, context-complete forms that are easier for retrievers to process accurately** - rewriting resolves ambiguity, references, and noisy phrasing before search. **What Is Query rewriting?** - **Definition**: Reformulation step that preserves intent while improving retrievability. - **Common Rewrites**: Coreference resolution, spelling normalization, explicit entity insertion, and intent clarification. - **Dialogue Use Case**: Converts follow-up questions into standalone retrieval-ready queries. - **Method Options**: Rule-based rewriting, sequence models, or LLM-based rewrite agents. **Why Query rewriting Matters** - **Retrieval Precision**: Cleaner, explicit queries improve first-stage candidate relevance. - **Conversation Support**: Handles pronouns and implicit references in multi-turn chat. - **Noise Reduction**: Removes irrelevant conversational fillers that confuse search. - **Latency Savings**: Better initial query reduces repeated retrieval retries. - **Answer Quality**: Stronger evidence selection improves final grounded responses. **How It Is Used in Practice** - **Rewrite Constraints**: Preserve user intent and avoid introducing unsupported assumptions. - **Quality Checks**: Validate rewrite equivalence before retrieval execution. - **Fallback Strategy**: Run both original and rewritten queries when confidence is low. Query rewriting is **a high-impact pre-retrieval optimization for RAG** - intent-preserving reformulation substantially improves evidence retrieval and downstream answer reliability in conversational settings.