One Time of Interaction May Not Be Enough: Go Deep with an Interaction-over-Interaction Network for Response Selection in Dialogues

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1–11 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 1 One Time of Interaction May Not Be Enough: Go Deep with an Interaction-over-Interaction Network for Response Selection in Dialogues Chongyang Tao1, Wei Wu2, Can Xu2, Wenpeng Hu1, Dongyan Zhao1,3 and Rui Yan1,3∗ 1Institute of Computer Science and Technology, Peking University, Beijing, China 2Microsoft Corporation, Beijing, China 3Center for Data Science, Peking University, Beijing, China 1,3{chongyangtao,wenpeng.hu,zhaody,ruiyan}@pku.edu.cn 2{wuwei,caxu}@microsoft.com Abstract Currently, researchers have paid great at- tention to retrieval-based dialogues in open- domain. In particular, people study the prob- lem by investigating context-response match- ing for multi-turn response selection based on publicly recognized benchmark data sets. State-of-the-art methods require a response to interact with each utterance in a context from the beginning, but the interaction is per- formed in a shallow way. In this work, we let utterance-response interaction go deep by proposing an interaction-over-interaction network (IoI). The model performs match- ing by stacking multiple interaction blocks in which residual information from one time of interaction initiates the interaction process again. Thus, matching information within an utterance-response pair is extracted from the interaction of the pair in an iterative fashion, and the information flows along the chain of the blocks via representations. Evaluation re- sults on three benchmark data sets indicate that IoI can significantly outperform state-of-the- art methods in terms of various matching met- rics. Through further analysis, we also unveil how the depth of interaction affects the perfor- mance of IoI. 1 Introduction Building a chitchat style dialogue systems in open- domain for human-machine conversations has at- tracted increasing attention in the conversational artificial intelligence (AI) community. Generally speaking, there are two approaches to implement- ing such a conversational system. The first ap- proach leverages techniques of information re- trieval (Lowe et al., 2015; Wu et al., 2017; Yan and Zhao, 2018), and selects a proper response from an index; while the second approach di- rectly synthesizes a response with a natural lan- ∗Corresponding author: Rui Yan (ruiyan@pku.edu.cn). guage generation model estimated from a large- scale conversation corpus (Serban et al., 2016; Li et al., 2017b). In this work, we study the prob- lem of multi-turn response selection for retrieval- based dialogue systems where the input is a con- versation context consisting of a sequence of utter- ances. Compared with generation-based methods, retrieval-based methods are superior in terms of response fluency and diversity, and thus have been widely applied in commercial chatbots such as the social bot XiaoIce (Shum et al., 2018) from Mi- crosoft, and the e-commerce assistant AliMe As- sist from Alibaba Group (Li et al., 2017a). A key step in multi-turn response selection is to measure the matching degree between a con- versation context and a response candidate. State- of-the-art methods (Wu et al., 2017; Zhou et al., 2018b) perform matching within a representation- interaction-aggregation framework (Wu et al., 2018b) where matching signals in each utterance- response pair are distilled …

Content truncated to 500 words