A Sequential Flow Control Framework for Multi-hop Knowledge Base Question Answering

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 8450 - 8460 December 7-11, 2022 ©2022 Association for Computational Linguistics A Sequential Flow Control Framework for Multi-hop Knowledge Base Question Answering Minghui Xie, Chuzhan Hao, and Peng Zhang∗ College of Intelligence and Computing, Tianjin University {minghuixie, chuzhanhao, pzhang}@tju.edu.cn Abstract One of the key challenges of knowledge base question answering (KBQA) is the multi-hop reasoning. Since in different hops, one attends to different parts of question, it is important to dynamically represent the question seman- tics for each hop. Existing methods, however, (i) infer the dynamic question representation only through coarse-grained attention mecha- nisms, which may bring information loss, (ii) and have not effectively modeled the sequen- tial logic, which is crucial for the multi-hop reasoning process in KBQA. To address these issues, we propose a sequential reasoning self- attention mechanism to capture the crucial rea- soning information of each single hop in a more fine-grained way. Based on Gated Re- current Unit (GRU) which is good at modeling sequential process, we propose a simple but effective GRU-inspired Flow Control (GFC) framework to model sequential logic in the whole multi-hop process. Extensive experi- ments on three popular benchmark datasets have demonstrated the superior effectiveness of our model. In particular, GFC achieves new state-of-the-art Hits@1 of 76.8% on WebQSP and is also effective when KB is incomplete. Our code and data are available at https: //github.com/Xie-Minghui/GFC. 1 Introduction Knowledge base question answering (KBQA) aims to answer questions from structured knowledge bases. In real application scenarios of KBQA, rea- soning with multiple hops over knowledge graph (KG) is necessary for answering complex questions. Therefore, how to perform multi-hop reasoning ef- fectively becomes a key challenge for multi-hop KBQA task (Sun et al., 2018; Zhang et al., 2018; Ho et al., 2020; Shi et al., 2020; Han et al., 2020). Existing methods for multi-hop KBQA have three main strands. The first is semantic parsing ∗Corresponding author. GFC 0.964 Predicted Entity Auxiliary Entity hop 1 hop 2 TransferNet Question: What character did john noble play in lord of the rings? feedback inhibition Relation Attention Weight e1 0.664 e0 e1 0.997 e2 0.858 r1 e1 0.952 e0 e1 0.949 e2 e0 r1(film.actor.film) r2 (film.performance.character) r1 r1 r1 r2 r2 Path: e0 (John Noble) e1 (The Lord of the Rings: e2 (Ans: Denethor II) e0 The Two Towers) Figure 1: The above picture shows relations attention weights on the reasoning paths of GFC and the strong path-based method TransferNet. The final entity scores are the weighted sum of two hops which are positive correlation with relation attention weights. TransferNet tends to give r1 high score in the 2nd hop, thus obtaining wrong answer (right). GFC can effectively weaken the attention of r1 in the 2nd hop by introducing GRU- like sequential logic into the multi-hop process (left). People tend to pay more attention to current relations while pay less attention to past relations. Thus GFC is more consistent with human reasoning habit. based …

Content truncated to 500 words