Fact-Tree Reasoning for N-ary Question Answering over Knowledge Graphs

Findings of the Association for Computational Linguistics: ACL 2022, pages 788 - 802 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Fact-Tree Reasoning for N-ary Question Answering over Knowledge Graphs Yao Zhang1 Peiyao Li1 Hongru Liang2 Adam Jatowt3 Zhenglu Yang1∗ 1TKLNDST, CS, Nankai University, China, 2Sichuan University, China, 3University of Innsbruck, Austria {yaozhang, peiyao_li}@mail.nankai.edu.cn, lianghongru@scu.edu.cn, adam.jatowt@uibk.ac.at, yangzl@nankai.edu.cn Abstract Current Question Answering over Knowledge Graphs (KGQA) task mainly focuses on per- forming answer reasoning upon KGs with bi- nary facts. However, it neglects the n-ary facts, which contain more than two entities. In this work, we highlight a more challenging but under-explored task: n-ary KGQA, i.e., an- swering n-ary facts questions upon n-ary KGs. Nevertheless, the multi-hop reasoning frame- work popular in binary KGQA task is not di- rectly applicable on n-ary KGQA. We propose two feasible improvements: 1) upgrade the ba- sic reasoning unit from entity or relation to fact, and 2) upgrade the reasoning structure from chain to tree. Therefore, we propose a novel fact-tree reasoning framework, FacTree, which integrates the above two upgrades. Fac- Tree transforms the question into a fact tree and performs iterative fact reasoning on the fact tree to infer the correct answer. Experi- mental results on the n-ary KGQA dataset we constructed and two binary KGQA benchmarks demonstrate the effectiveness of FacTree com- pared with state-of-the-art methods. 1 Introduction The task of Question Answering over Knowl- edge Graphs (KGQA) has provided new av- enues to the recent development of QA sys- tems by utilizing the advantages of KGs (Yu et al., 2017; Dubey et al., 2019; Huang et al., 2019; Zhang et al., 2021). Current KGQA stud- ies mainly consider performing answer reason- ing upon KGs with binary facts, which encode binary relations between pairs of entities, e.g., Golden State Warriors’ arena is Chase Center1. However, n-ary facts that involve more than two entities are also ubiquitous in reality (Guan et al., 2019; Abboud et al., 2020; Wang et al., 2021), e.g., the ternary fact Golden State Warriors won ∗Corresponding author. 1Entities are underlined. the NBA championship in 2018. Compared to bi- nary facts, n-ary facts have more information con- tent. This makes the answer reasoning for ques- tions involving n-ary facts more intractable, ex- posing open challenges in KGQA. In this work, we aim to study the under-explored n-ary KGQA task, i.e., answering n-ary facts questions upon n-ary KGs. The multi-hop reasoning KGQA method (Das et al., 2018; Qiu et al., 2020; Saxena et al., 2020; Ren et al., 2021) has become popular for its high efficiency and interpretability. Specifically, the rea- soning process can be expressed as a chain, start- ing from an entity extracted from the question and then walking on the KG by connected relations and entities until arriving at the answer entity. See Fig- ure 1 (b) for an example, to answer the question, what is the address of the arena of the Golden State Warriors, the reasoning chain starts from Golden State Warriors, to walk through “arena→Chase Center→address", …

Content truncated to 500 words