ASIM: Explicit Slot-Intent Mapping with Attention for Joint Multi-intent Detection and Slot Filling

Jingwen Chen, Xingyi Liu, Mingzhi Wang, Yongli Xiao, Dianhui Chu, Chunshan Li, Feng Wang

Abstract: The accurate analysis of a user’s natural language statement, including their potential intentions and corresponding slot tags, is crucial for cognitive intelligence services. In real-world applications, a user’s statement often contains multiple intentions, and most existing approaches either mainly focus on the single-intent research problems or utilizes an overall encoder directly to capture the relationship between intents and slot tags, which ignore the explicit slot-intent mapping relation. In this paper, we propose a novel Attention-based Slot-Intent Mapping Method (ASIM) for joint multi-intent detection and slot filling task. The ASIM model not only models the correlation among sequence tags while considering the mutual influence between two tasks but also maps specific intents to each semantic slot. The ASIM model can balance multi-intent knowledge to guide slot filling and further increase the interaction between the two tasks. Experimental results on the MixATIS dataset demonstrate that our ASIM model achieves substantial improvement and state-of-the-art performance.