燕山大学学报2025，Vol.49Issue(6)：525-537,543,14.DOI:10.3969/j.issn.1007-791X.2025.06.007
一种基于多级图注意力的意图识别和槽位填充方法
An intent recognition and slot filling method based on multi-level graph attention
摘要
Abstract
The current researches on intent recognition and slot filling are mainly based on unambiguous utterances.However,existing methods for unambiguous utterances perform poorly when handling ambiguous utterances in diverse contexts.An Intent recognition and Slot filling method based on Multi-level Graph Attention(ISMGA)is proposed,leveraging profile information to reduce the impact of polysemy and ambiguous expressions on intent recognition and slot filling.The ISMGA consists of a Sentence-Level Adaptive Graph Attention Module(SLAGAM)and a Word-Level Adaptive Graph Attention Module(WLAGAM).The SLAGAM integrates profile information and slot information into sentence-level representations to mitigate the adverse effects of ambiguity on intent recognition.The WLAGAM incorporates profile information into word-level representations to reduce the impact of ambiguity and explicitly uses intent information to guide slot filling.Experimental results on the Chinese dataset ProSLU,the self-built dataset FamSLU,and three unambiguous datasets CAIS,SMP2019,and SMP2020,demonstrate that ISMGA significantly improves performance and exhibits strong generalization ability.关键词
口语歧义/意图识别/槽位填充/图注意力网络Key words
spoken language ambiguity/intent recognition/slot filling/graph attention network分类
信息技术与安全科学引用本文复制引用
李贤善,王森,师晓倩,赵逢达..一种基于多级图注意力的意图识别和槽位填充方法[J].燕山大学学报,2025,49(6):525-537,543,14.基金项目
中央引导地方科技发展资金项目(246Z1817G) （246Z1817G）
新疆维吾尔自治区自然科学基金资助项目(2022D01A59) （2022D01A59）
河北省创新能力提升计划项目(22567637H) （22567637H）
