Multi-granularity Temporal Question Answering over Knowledge Graphs

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 11378–11392 July 9-14, 2023 ©2023 Association for Computational Linguistics Multi-granularity Temporal Question Answering over Knowledge Graphs Ziyang Chen1, Jinzhi Liao2, Xiang Zhao1,∗ 1 Laboratory for Big Data and Decision, National University of Defense Technology, China 2 National Defense University, China {chenziyangnudt, liaojinzhi12, xiangzhao}@nudt.edu.cn Abstract Recently, question answering over temporal knowledge graphs (i.e., TKGQA) has been in- troduced and investigated, in quest of reasoning about dynamic factual knowledge. To foster re- search on TKGQA, a few datasets have been curated (e.g., CRONQUESTIONS and Complex- CRONQUESTIONS), and various models have been proposed based on these datasets. Never- theless, existing efforts overlook the fact that real-life applications of TKGQA also tend to be complex in temporal granularity, i.e., the ques- tions may concern mixed temporal granularities (e.g., both day and month). To overcome the limitation, in this paper, we motivate the notion of multi-granularity temporal question answer- ing over knowledge graphs and present a large- scale dataset for multi-granularity TKGQA, namely MULTITQ. To the best of our knowl- edge, MULTITQ is among the first of its kind, and compared with existing datasets on TKGQA, MULTITQ features at least two desir- able aspects—ample relevant facts and multiple temporal granularities. It is expected to better reflect real-world challenges, and serve as a test bed for TKGQA models. In addition, we propose a competing baseline MultiQA over MULTITQ, which is experimentally demon- strated to be effective in dealing with TKGQA. The data and code are released at https: //github.com/czy1999/MultiTQ. 1 Introduction In real-life applications factual knowledge is apt to evolve over time (Nonaka et al., 2000; Roddick and Spiliopoulou, 2002; Hoffart et al., 2011; Gottschalk and Demidova, 2018); for instance, The host city of the Winter Olympic Games in 2018 was South Ko- rea, while in 2022 it was Beijing.In this connection, there is a current trend to investigate knowledge graphs (KGs) involving time, and these KGs are coined as temporal knowledge graphs (TKGs). In a ∗Corresponding author. Figure 1: Examples of temporal question answering. TKG, fact triplets are equipped with temporal infor- mation (e.g., timestamps), and a temporal fact can be stated in the form like “(Beijing, held, Winter Olympic Games, 2022)”. To exploit the value of TKGs, recent research effort has been devoted to process natural language questions over TKG, i.e., question answering over TKG (TKGQA in short) (Saxena et al., 2021). Given a question and a background TKG, it re- trieves from the TKG an answer to the question. To foster research on TKGQA, several datasets have been introduced, among which CRONQUES- TIONS (Saxena et al., 2021) is by far the largest. We explain the task with a sample question in CRON- QUESTIONS. Example 1 In the upper part of Figure 1, the agent is supplied with the question “What award did Carlo Taverna receive in 1863?” as well as a TKG. By considering the semantic relevance of the facts, the agent locates the candidate fact …

Content truncated to 500 words