Research on Clarification Question Recognition and Generation in Intelligent Q&A

Intelligent question answering systems often encounter ambiguous questions that require the generation of clarification questions to understand users’ true intentions. Without clarification questions, syestems may be confused by ambiguous questions. In this paper, the generation of clarification questions is separated into three subtasks. This study focuses on the three subtasks of the clarification question identification and generation process. We propose the DeBERTA v3 + FC model for clarification question detection and entity prediction, and an improved ByT5-based model for generating diverse and comprehensible clarification questions. On the MSParS dataset, our method outperforms traditional DMN models by 11.2% and 15.17% in accuracy for clarification question detection and entity prediction tasks respectively, while the BLEU score is 4.9% higher than the traditional Seq2Seq models. The efficacy of our proposed methods is verified by their superior performance to traditional methods on all three subtasks. The results on the MSParS dataset demonstrate the effectiveness of our framework for generating clarification questions and entity predictions.