GMH: A General Multi-hop Reasoning Model for KG Completion

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 3437–3446 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 3437 GMH: A General Multi-hop Reasoning Model for KG Completion Yao Zhang1 Hongru Liang2 Adam Jatowt3 Wenqiang Lei4 Xin Wei5 Ning Jiang5 Zhenglu Yang1∗ 1TKLNDST, CS, Nankai University, China, 2Sichuan University, China, 3University of Innsbruck, Austria, 4National University of Singapore, Singapore, 5Mashang Consumer Finance Co., Ltd., China yaozhang@mail.nankai.edu.cn, lianghongru@scu.edu.cn, adam.jatowt@uibk.ac.at, wenqianglei@gmail.com, {xin.wei02, ning.jiang02}@msxf.com, yangzl@nankai.edu.cn Abstract Knowledge graphs are essential for numerous downstream natural language processing ap- plications, but are typically incomplete with many facts missing. This results in research efforts on multi-hop reasoning task, which can be formulated as a search process and current models typically perform short distance rea- soning. However, the long-distance reasoning is also vital with the ability to connect the su- perficially unrelated entities. To the best of our knowledge, there lacks a general framework that approaches multi-hop reasoning in mixed long-short distance reasoning scenarios. We argue that there are two key issues for a gen- eral multi-hop reasoning model: i) where to go, and ii) when to stop. Therefore, we pro- pose a general model which resolves the is- sues with three modules: 1) the local-global knowledge module to estimate the possible paths, 2) the differentiated action dropout mod- ule to explore a diverse set of paths, and 3) the adaptive stopping search module to avoid over searching. The comprehensive results on three datasets demonstrate the superiority of our model with significant improvements against baselines in both short and long dis- tance reasoning scenarios. 1 Introduction Knowledge graphs (KGs) have become the pre- ferred technology for representing, sharing and adding factual knowledge to many natural lan- guage processing applications like recommenda- tion (Wang et al., 2019; Lei et al., 2020) and ques- tion answering (Huang et al., 2019; Zhang et al., 2018). KGs store triple facts (head entity, relation, tail entity) in the form of graphs, where entities are represented as nodes and relations are represented as labeled edges between entities (e.g., Figure 1 (a)). Although popular KGs already contain mil- lions of facts, e.g., YAGO (Suchanek et al., 2007) ∗Corresponding author. and Freebase (Bollacker et al., 2008), they are far from being complete considering the amount of ex- isting facts and the scope of continuously appearing new knowledge. This has become the performance bottleneck of many KG-related applications, trig- gering research efforts on the multi-hop reasoning task. The multi-hop reasoning task can be formu- lated as a search process, in which the search agent traverses a logical multi-hop path to find the missing tail entity of an incomplete triple in KG. As shown in Figure 1 (b), the two-hop path Stephen Curry plays for −−−−−−→ Golden State War- riors plays for ←−−−−−− Klay Thompson is searched to reason Klay Thompson as the missing entity of (Stephen Curry, teammate, ⋆), where ⋆ denotes the missing tail entity. Multi-hop reasoning meth- ods (Xiong et al., 2017; Das et al., 2018) have been proposed to …

Content truncated to 500 words