DUTIR at SemEval-2026 Task 8: A Hybrid Retrieval and Faithfulness-Guarded Framework for Multi-Turn RAG
Year: 2026
Type: conference-paper
Abstract: This paper describes the system submitted by DUTIR_taskC for SemEval-2026 Task 8: MTRAGEval (Task C).Multi-turn Retrieval-Augmented Generation (RAG) poses significant challenges in context tracking, retrieval precision, and hallucination mitigation.Our proposed system addresses these by employing a multi-stage pipeline consisting of: (1) LLMbased query rewriting (powered by GPT-5.2) to resolve conversational dependencies; (2) a hybrid retrieval module combining dense embeddings (BGE-M3) and sparse retrieval (BM25) with Reciprocal Rank Fusion (RRF); (3) a confidence-based answerability gating mechanism; and (4) a post-generation faithfulness guard.Experimental results on the blind test set show that our approach achieves a Composite Score of 0.5576, ranking 4th out of 29 participating teams.Detailed analysis reveals that our system significantly outperforms strong baselines in faithfulness and successfully handles underspecified queries.
Authors: Jin Yang, Yichong Chen, Liang Yang
Institution: Dalian University of Technology
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Funder: Dalian University of Technology
OpenAlex ID: W7166867393