A Small Model Asks Better: Fine-Tuning Query Clarification for Retrieval-Augmented Generation

Retrieval‑augmented generation (RAG) systems rely on clear user queries, because ambiguous prompts can make it hard to pick relevant documents. In multi‑turn conversational settings, a question may be poorly phrased, confusing within prior context, or well written yet open to multiple interpretations when compared to retrieved texts. The goal is to spot both cases and ask the user for clarification. To tackle this, researchers introduce ClariEdge, a dataset with roughly 3,900 examples of ambiguous questions and negative cases that do not need clarification. They show that a small model fine‑tuned with LoRA achieves a classification accuracy of 96.6 % and outperforms several larger prompt‑based models on both classification and the generation of high‑quality clarifying requests. The work emphasizes the value of hard negative examples to distinguish truly ambiguous queries from those resolvable with available context. The same LoRA model works effectively both before and after the retrieval step, enabling its integration at two distinct points in the RAG pipeline as a pre‑generation filter. This research suggests that enhancing a system’s ability to seek clarifications can raise the precision and usefulness of generated answers, which is important for applications that depend on natural user interaction.

Read the original article on arXiv cs.AI

This summary is an informational synthesis produced by dataqbs.com. All rights to the original content belong to its author and the cited media outlet. We act solely as curators of technology news and claim no authorship.