Sequential Matching Network: A New Architecture for Multi-turn Response Selection in Retrieval-Based Chatbots

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 496–505 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1046 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 496–505 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1046 Sequential Matching Network: A New Architecture for Multi-turn Response Selection in Retrieval-Based Chatbots Yu Wu†, Wei Wu‡, Chen Xing♦, Zhoujun Li†∗, Ming Zhou‡ †State Key Lab of Software Development Environment, Beihang University, Beijing, China ♦College of Computer and Control Engineering, Nankai University, Tianjin, China ‡ Microsoft Research, Beijing, China {wuyu,lizj}@buaa.edu.cn {wuwei,v-chxing,mingzhou}@microsoft.com Abstract We study response selection for multi- turn conversation in retrieval-based chat- bots. Existing work either concatenates ut- terances in context or matches a response with a highly abstract context vector fi- nally, which may lose relationships among utterances or important contextual infor- mation. We propose a sequential match- ing network (SMN) to address both prob- lems. SMN first matches a response with each utterance in the context on multiple levels of granularity, and distills important matching information from each pair as a vector with convolution and pooling oper- ations. The vectors are then accumulated in a chronological order through a recur- rent neural network (RNN) which models relationships among utterances. The fi- nal matching score is calculated with the hidden states of the RNN. An empirical study on two public data sets shows that SMN can significantly outperform state- of-the-art methods for response selection in multi-turn conversation. 1 Introduction Conversational agents include task-oriented dia- log systems and non-task-oriented chatbots. Dia- log systems focus on helping people complete spe- cific tasks in vertical domains (Young et al., 2010), while chatbots aim to naturally and meaningfully converse with humans on open domain topics (Ritter et al., 2011). Existing work on building chatbots includes generation -based methods and retrieval-based methods. Retrieval based chatbots enjoy the advantage of informative and fluent re- sponses, because they select a proper response for ∗ Corresponding Author Context utterance 1 Human: How are you doing? utterance 2 ChatBot: I am going to hold a drum class in Shanghai. Anyone wants to join? The location is near Lujiazui. utterance 3 Human: Interesting! Do you have coaches who can help me practice drum? utterance 4 ChatBot: Of course. utterance 5 Human: Can I have a free first lesson? Response Candidates response 1 Sure. Have you ever played drum before? ✓ response 2 What lessons do you want?  Table 1: An example of multi-turn conversation the current conversation from a repository with re- sponse selection algorithms. While most existing work on retrieval-based chatbots studies response selection for single-turn conversation (Wang et al., 2013) which only considers the last input message, we consider the problem in a multi-turn scenario. In a chatbot, multi-turn response selection takes a message and utterances in its previous turns as input and selects a response that is natural and rel- evant to the whole context. The key to …

Content truncated to 500 words