Task-Oriented Conversation Generation Using Heterogeneous Memory Networks

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 4558–4567, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 4558 Task-Oriented Conversation Generation Using Heterogeneous Memory Networks Zehao Lin 1 Xinjing Huang 1 Feng Ji 2 Haiqing Chen 2 Yin Zhang1∗ 1 College of Computer Science and Technology, Zhejiang University 2 DAMO Academy, Alibaba Group {georgelin,huangxinjing,zhangyin98}@zju.edu.cn {zhongxiu.jf,haiqing.chenhq}@alibaba-inc.com Abstract How to incorporate external knowledge into a neural dialogue model is critically important for dialogue systems to behave like real hu- mans. To handle this problem, memory net- works are usually a great choice and a promis- ing way. However, existing memory networks do not perform well when leveraging hetero- geneous information from different sources. In this paper, we propose a novel and ver- satile external memory networks called Het- erogeneous Memory Networks (HMNs), to si- multaneously utilize user utterances, dialogue history and background knowledge tuples. In our method, historical sequential dialogues are encoded and stored into the context-aware memory enhanced by gating mechanism while grounding knowledge tuples are encoded and stored into the context-free memory. During decoding, the decoder augmented with HMNs recurrently selects each word in one response utterance from these two memories and a gen- eral vocabulary. Experimental results on mul- tiple real-world datasets show that HMNs sig- nificantly outperform the state-of-the-art data- driven task-oriented dialogue models in most domains. 1 Introduction Compared with chitchat, task-oriented dialogue systems aim at solving tasks in specific domains with grounding knowledge. Though far from han- dling conversation like a real human, existing task- oriented dialogue systems have shown cheerful prospect in a specific domain, e.g. Siri and Cor- tana are personal assistants, helping people a lot in daily life and business work. In general, knowledge-grounded task-oriented dialogue system can be divided into three impor- tant components: understanding user utterances, fetching right knowledge from external storage ∗*Corresponding Author Figure 1: A multi-turn dialogue example. The upper table shows several n-tuples sampled from knowledge base. Lower table shows multi-turn dialogues. Agent needs to retrieve appropriate knowledge tuples to gen- erate the proper response. and replying right answer. As shown in Fig- ure 1, agent is required to do a point-of-interest navigation. According to dialogue history, agent will fetch related knowledge base information, in our case represented as tuples (e.g. [hotel keen, poi type, rest stop], which indicates the point- of-i nterests type of hotel keen is rest stop), as an external knowledge to answer correctly and com- plete task. Traditional pipeline dialogue systems(Yan et al., 2017; Rojas-Barahona et al., 2017) and some end-to-end dialogue systems rely on the predefined the slot filling labels. Despite the consumption of human efforts, these kinds of systems are difficult to adapt to new domains. Many works, e.g. (Vinyals and Le, 2015; Shang et al., 2015), show that training a fully data-driven end-to-end model is a promising way to build domain-agnostic dialogue system. Their mod- els mostly try to use the attention mechanism, …

Content truncated to 500 words