Filtering before Iteratively Referring for Knowledge-Grounded Response Selection in Retrieval-Based Chatbots

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1412–1422 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 1412 Filtering before Iteratively Referring for Knowledge-Grounded Response Selection in Retrieval-Based Chatbots Jia-Chen Gu1, Zhen-Hua Ling1∗, Quan Liu1,2, Zhigang Chen2, Xiaodan Zhu3 1National Engineering Laboratory for Speech and Language Information Processing, University of Science and Technology of China, Hefei, China 2State Key Laboratory of Cognitive Intelligence, iFLYTEK Research, Hefei, China 3ECE & Ingenuity Labs, Queen’s University, Kingston, Canada gujc@mail.ustc.edu.cn, {zhling, quanliu}@ustc.edu.cn, zgchen@iflytek.com, xiaodan.zhu@queensu.ca Abstract The challenges of building knowledge- grounded retrieval-based chatbots lie in how to ground a conversation on its background knowledge and how to match response candidates with both context and knowledge simultaneously. This paper proposes a method named Filtering before Iteratively REferring (FIRE) for this task. In this method, a context filter and a knowledge filter are first built, which derive knowledge-aware context representations and context-aware knowledge representations respectively by global and bidirectional attention. Besides, the entries irrelevant to the conversation are discarded by the knowledge filter. After that, iteratively referring is performed between context and response representations as well as between knowledge and response representations, in order to collect deep matching features for scoring response candidates. Experimental results show that FIRE outperforms previous methods by margins larger than 2.8% and 4.1% on the PERSONA-CHAT dataset with original and revised personas respectively, and margins larger than 3.1% on the CMU DoG dataset in terms of top-1 accuracy. We also show that FIRE is more interpretable by visualizing the knowledge grounding process. 1 Introduction Building a conversational agent with intelligence has received significant attention with the emer- gence of personal assistants such as Apple Siri, Google Now and Microsoft Cortana. One approach is to building retrieval-based chatbots, which aims to select a potential response from a set of candi- dates given the conversation context (Lowe et al., 2015; Wu et al., 2017; Zhou et al., 2018b; Gu et al., 2019a; Tao et al., 2019; Gu et al., 2020a). ∗Corresponding author. The inception 2009 Christopher Nolan Scientific Leonardo DiCaprio as Dom Cobb, a professional thief who specializes in conning secrets from his victims by infiltrating their dreams. Tom Hardy as Eames, a sharp-tongued associate of Cobb. ... Response DiCaprio, who has never been better as the tortured hero, draws you in with a love story that will appeal even to non-scifi fans. The movie is a metaphor for the power of delusional hype for itself. ... Dominick Cobb and Arthur are extractors, who perform corporate espionage using an experimental military technology to infiltrate the subconscious of their targets and extract valuable information through a shared dream world. Their latest target, Japanese businessman Saito, reveals that he arranged the mission himself to test Cobb for a seemingly impossible job: planting an idea in a person’s subconscious, or inception. Rotten Tomatoes: 86% and average: 8.1/10; IMDB: 8.8/10 Name Year Director Genre Cast Critical Response Introd- -uction Rating Background Knowledge Conversation Hi how are you today? I am good. How are …

Content truncated to 500 words