History Semantic Graph Enhanced Conversational KBQA with Temporal Information Modeling

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 3521–3533 July 9-14, 2023 ©2023 Association for Computational Linguistics History Semantic Graph Enhanced Conversational KBQA with Temporal Information Modeling Hao Sun1, Yang Li2, Liwei Deng3, Bowen Li2, Binyuan Hui2 Binhua Li2, Yunshi Lan4, Yan Zhang1, Yongbin Li 2 1 Peking University, 2 Alibaba Group 3 University of Electronic Science and Technology of China, 4 East China Normal University sunhao@stu.pku.edu.cn, zhyzhy001@pku.edu.cn {ly200170, binyuan.hby, binhua.lbh, shuide.lyb}@alibaba-inc.com deng_liwei@std.uestc.edu.cn, libowen.ne@gmail.com, yslan@dase.ecnu.edu.cn Abstract Context information modeling is an important task in conversational KBQA. However, ex- isting methods usually assume the indepen- dence of utterances and model them in isolation. In this paper, we propose a History Semantic Graph Enhanced KBQA model (HSGE) that is able to effectively model long-range se- mantic dependencies in conversation history while maintaining low computational cost. The framework incorporates a context-aware en- coder, which employs a dynamic memory de- cay mechanism and models context at different levels of granularity. We evaluate HSGE on a widely used benchmark dataset for complex sequential question answering. Experimental results demonstrate that it outperforms existing baselines averaged on all question types. 1 Introduction In recent years, with the development of large-scale knowledge base (KB) like DBPedia (Auer et al., 2007) and Freebase (Bollacker et al., 2008), Knowl- edge Base Question Answering (KBQA) (Wang et al., 2020; Ye et al., 2021; Yan et al., 2021; Yadati et al., 2021; Das et al., 2021; Wang et al., 2022) has become a popular research topic, which aims to convert a natural language question to a query over a knowledge graph to retrieve the correct an- swer. With the increasing popularity of AI-driven assistants (e.g., Siri, Alexa and Cortana), research focus has shifted towards conversational KBQA (Shen et al., 2019; Kacupaj et al., 2021; Marion et al., 2021) that involves multi-turn dialogues. A common solution to the task of conversational KBQA is to map an utterance to a logical form using semantic parsing approach (Shen et al., 2019; Guo et al., 2018). The state-of-the-art semantic parsing approach (Kacupaj et al., 2021) breaks down the process into two stages: a logical form is first generated by low-level features and then the missing details are filled by taking both the Q1: Who is the president of the united states? R1: Joe Biden Q2: Where did he graduate from? R2: University of Delaware Q3: When was it established? R3: 1743 Q4: How about Harvard university? R4: 1636 Figure 1: An example illustrating the task of conversa- tional KBQA. question and templates into consideration. Other approaches (Dong and Lapata, 2016; Liang et al., 2016; Guo et al., 2018) mainly focus on first detect- ing entities in the question and then mapping the question to a logical form. Despite the inspiring results of the semantic pars- ing methods mentioned above, most of them fail to model the long-range semantic dependency in conversation history. Specifically, they usually di- rectly incorporate immediate two turns of conversa- tions and ignore the …

Content truncated to 500 words