Historical Information-Based Intent Detection for Multiturn Dialogue
Xin Sun, Hongchao Zheng, Zheng Tang · 2022
Intent detection aims to determine the intent of users, an important task in natural language processing and dialogue systems. As one of the key modules of task-based dialogue systems, intent detection directly influences the meaning analysis of spoken language and the performance of dialogue systems. However, existing intent detection algorithms mainly focus on single-turn conversation, ignoring the valuable historical conversation message. To alleviate the problem, we propose a multiturn dialogue intent detection model based on historical information. Specifically, the semantic information of user statements and system statements is extracted by a role-distinguished self-attention network. Then, semantic relevance based on context information adjustment is used to combine the system statement and user statement. Finally, long short-term memory (LSTM) is used to capture the historical message. Experimental results prove that the proposed method improves the effect of intent detection and performs better than seven mainstream baselines.