Faithful queries improve conversational search with balanced relevance
Faithful Query Resolution for Conversational Search
Information Retrieval
Summary
When you search using a conversation, sometimes the system adds extra meaning you didn't say, which can change the results. The authors studied how staying true to the user's exact words affects what information is found. They tested methods that keep queries faithful while also mixing in expanded searches to find more results. Their approach improved how many relevant results show up, but didn't always make users happier with the conversation. This suggests it's useful to separate faithful understanding of queries from broader searching when building conversational search tools.
What this means in practice
- •For search engine developers: Enhance conversational assistants by combining faithful query rewriting with retrieval expansion to improve relevant document recall.
- •For customer support software teams: Improve chat-based help systems by distinguishing query intent faithfully from expanded search to boost answer relevance without harming user satisfaction.
Authors
Saber Zerhoudi, Ivica Kostric, Jelena Mitrovic
Abstract
Conversational query rewriting makes context-dependent requests explicit for retrieval, but additional context can introduce an information need the user did not express. This raises a question about the interaction between query fidelity and retrieval breadth. This paper examines that interaction in behavior-adaptive conversational search, using iKAT SCAI 2026 as the evaluation setting. We compare five query-generation configurations within a behavior-adaptive assistant using BM25 and SPLADE retrieval. A minimal-faithful resolver preserves the user's wording and resolves only context-dependent references; weighted reciprocal rank fusion combines its rankings with those from a learned rewriter and the raw utterance. Offline evaluation shows that adding these formulations recovers recall over the faithful query alone, with a slight reduction in nDCG@10. Across our five iKAT SCAI 2026 submissions, the weighted configuration obtains the highest mean retrieval scores, reaching nDCG@5 of 0.534, a 5.3\% relative improvement over the strongest alternative. However, this retrieval advantage does not coincide with the highest conversational satisfaction scores. These findings motivate separating faithful query interpretation from retrieval expansion, while assessing their effects on conversational utility independently.