Intention Reasoning Network for Multi-Domain End-to-end Task-Oriented Dialogue

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 2273–2285 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 2273 Intention Reasoning Network for Multi-Domain End-to-end Task-Oriented Dialogue Zhiyuan Ma1, Jianjun Li1∗, Zezheng Zhang1, Guohui Li1, Yongjing Cheng2 1 Huazhong University of Science and Technology (HUST), China 2 National Unveristy of Defense Technology (NUDT), China {zhiyuanma,jianjunli,zezhengzhang,guohuili}@hust.edu.cn davidcheng1001@163.com Abstract Recent years has witnessed the remarkable success in end-to-end task-oriented dialog sys- tem, especially when incorporating external knowledge information. However, the quality of most existing models’ generated response is still limited, mainly due to their lack of fine- grained reasoning on deterministic knowledge (w.r.t. conceptual tokens), which makes them difficult to capture the concept shifts and iden- tify user’s real intention in cross-task scenar- ios. To address these issues, we propose a novel intention mechanism to better model de- terministic entity knowledge. Based on such a mechanism, we further propose an inten- tion reasoning network (IR-Net), which con- sists of joint and multi-hop reasoning, to ob- tain intention-aware representations of concep- tual tokens that can be used to capture the con- cept shifts involved in task-oriented conversa- tions, so as to effectively identify user’s in- tention and generate more accurate responses. Experimental results verify the effectiveness of IR-Net, showing that it achieves the state- of-the-art performance on two representative multi-domain dialog datasets. 1 Introduction Task-oriented dialogue systems are designed to help users to achieve specific goals such as schedule ar- rangement or weather inquiry via natural language. Compared with traditional pipeline dialogue sys- tems (Young et al., 2013) that include multiple mod- ules each requiring a huge amount of human effort to design, end-to-end approaches (Gülçehre et al., 2016; Wen et al., 2017; Eric and Manning, 2017; Eric et al., 2017; Zhao et al., 2017; Quan et al., 2019; Moon et al., 2019; Jung et al., 2020; Dai et al., 2020) that can directly output system responses with plain text as input have recently gained much attention. In recent years, sequence to sequence (Seq2Seq) models have dominated the study of ∗Corresponding author. Figure 1: Example of task-oriented dialog including concept shifts from the SMD dataset (Eric et al., 2017). The solid arrows indicate the existed relationships be- tween entities, and the dotted arrows indicate the latent entity relationships (captured by IR-Net). The colored dotted boxes in the dialog indicate generated entities. end-to-end task-oriented dialog systems, and many memory augmented Seq2Seq models have been proposed (Bordes et al., 2017; Madotto et al., 2018; Wu et al., 2019; Wen et al., 2018; Qin et al., 2019; Reddy et al., 2019; Wang et al., 2020), which exploit both dialog history and domain-specific knowledge base (KB) to incorporate KB informa- tion and perform knowledge-based reasoning for better performance. Though achieving remarkable progress, existing memory augmented Seq2Seq models still suffer from the following two limitations. First, prior models rely heavily on the soft attention mecha- nism (Vaswani et al., 2017) to generate responses by adopting a weighted sum over the embeddings of memory triples …

Content truncated to 500 words