GraphDialog: Integrating Graph Knowledge into End-to-End Task-Oriented Dialogue Systems

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 1878–1888, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 1878 GraphDialog: Integrating Graph Knowledge into End-to-End Task-Oriented Dialogue Systems Shiquan Yang, Rui Zhang∗, Sarah Erfani The University of Melbourne, Australia {shiquan@student., rui.zhang@, sarah.erfani@}unimelb.edu.au Abstract End-to-end task-oriented dialogue systems aim to generate system responses directly from plain text inputs. There are two challenges for such systems: one is how to effectively incor- porate external knowledge bases (KBs) into the learning framework; the other is how to accurately capture the semantics of dialogue history. In this paper, we address these two challenges by exploiting the graph structural information in the knowledge base and in the dependency parsing tree of the dialogue. To ef- fectively leverage the structural information in dialogue history, we propose a new recurrent cell architecture which allows representation learning on graphs. To exploit the relations between entities in KBs, the model combines multi-hop reasoning ability based on the graph structure. Experimental results show that the proposed model achieves consistent improve- ment over state-of-the-art models on two dif- ferent task-oriented dialogue datasets. 1 Introduction Task-oriented dialogue systems aim to help user accomplish specific tasks via natural language inter- faces such as restaurant reservation, hotel booking and weather forecast. There are many commer- cial applications of this kind (e.g. Amazon Alexa, Google Home, and Apple Siri) which make our life more convenient. Figure 1 illustrates such an exam- ple where a customer is asking for the information about restaurants. By querying the knowledge base (KB), the agent aims to provide the correct restau- rant entities from the KB to satisfy the customer in a natural language form. Hence, the ability to understand the dialogue history, and to retrieve relevant information from the KB is essential in task-oriented dialogue systems. One approach for designing task-oriented dia- logue systems is the pipeline approach (Williams ∗Rui Zhang is the corresponding author. Turkish Chinese Golden Wok Efes Restaurant Moderate 30 King Street 106 Regent Street Food Food Price Price Address Address Customer: I’m looking for a moderately priced Polish restaurant. Agent: I’m sorry but there aren’t any Polish restaurants nearby. Customer: Can you please check for a Turkish restaurant? Agent: Efes Restaurant serves Turkish food with moderate price. Customer: May I have the address for Efes Restaurant? Agent: Yes, the address for Efes Restaurant is 30 King Street. Knowledge Base Dialogue Figure 1: An example dialogue in the restaurant book- ing domain. The top part is knowledge base (KB) infor- mation that represented by a graph and the bottom part is the conversation between a customer and the agent. Our aim is to predict the agent responses given KB in- formation and the customer utterances. and Young, 2007; Lee et al., 2009; Young et al., 2013), but it suffers from the difficulty in credit assignment and adaption to new domains. Another popular approach is the end-to-end models (Ser- ban et al., 2016; Wen et al., 2017; Williams et al., 2017; Zhao et …

Content truncated to 500 words