EDIT: Towards Enhancing Dialogue Response Generation for Large Language Models by Asking Questions to Detect User’s Intentions

Anonymous

Abstract: Large Language Models (LLMs), such as ChatGPT, have recently been applied to various NLP tasks due to its open-domain generation capabilities. However, during the dialogue process, users may have implicit intentions that might be overlooked by LLMs.Besides, it is unlikely for LLMs to encompass all fields comprehensively and LLMs cannot update the latest knowledge in real-time.To tackle these two issues, we propose a framework~\emph{using LLM to \textbf{E}nhance dialogue response generation by asking questions to \textbf{D}etect user's \textbf{I}mplicit in\textbf{T}entions} (\textbf{EDIT}). Firstly, we construct a \textit{Context-Open-Question} (COQ) dataset to train a question generator (QG) and generate open questions related to the dialogue context as the potential user's intention; Then, EDIT answers those questions by interacting with LLMs and retrieving domain-specific knowledge bases respectively; Finally, EDIT generates response by integrating those answers. To evaluate generated responses, we have specifically designed two metrics, \textit{Information Content} (IC) and \textit{Context Coherence} (CC), respectively.The results demonstrated significant improvements after combining current mainstream LLMs with EDIT on two task-oriented dialogue dataset (Wizard of Wikipedia and Holl-E).

Paper Type: long

Research Area: Dialogue and Interactive Systems

Contribution Types: NLP engineering experiment

Languages Studied: English