UHop: An Unrestricted-Hop Relation Extraction Framework for Knowledge-Based Question Answering

Abstract

In relation extraction for knowledge-based question answering, searching from one entity to another entity via a single relation is called "one hop". In related work, an exhaustive search from all one-hop relations, two-hop relations, and so on to the max-hop relations in the knowledge graph is necessary but expensive. Therefore, the number of hops is generally restricted to two or three. In this paper, we propose UHop, an unrestricted-hop framework which relaxes this restriction by use of a transition-based search framework to replace the relation-chain-based search one. We conduct experiments on conventional 1-and 2hop questions as well as lengthy questions, including datasets such as WebQSP, PathQuestion, and Grid World. Results show that the proposed framework enables the ability to halt, works well with state-of-the-art models, achieves competitive performance without exhaustive searches, and opens the performance gap for long relation paths.

Graph Convolution Network over Dependency Structure Improve Knowledge Base Question Answering

“…Reference [29] focused on multihop relations; however, without the limitation of the method, the search space grows exponentially. Chen et al (2019) [9] used a beam search to face the multihop questions, but it did not effectively handle the issue of constraints. We compared our method with that of Bhutani et al (2019) [30], which constructs complex query patterns using a set of simple queries.…”

“…Our method follows the same procedure as embedding-based models and regards query graph generation as a multistep relation path extraction process. References [9,10,24] proposed considering better relations. Lan et al (2020) [11] proposed modifying a query graph generation process from longer relations.…”

“…[8] 69.0 - [6] -40.9 [7] -42.8 [29] 67.9 - [9] 68.5 35.3 [30] 60.3 - [31] 72.6 - [11] 74.0 43.3 Our 74.8 44.2…”

“…To answer this type of question, the longer relation path has to be considered, which will increase the search space exponentially. The beam search method was introduced by References [9,10] to reduce the search space by considering the best matching relation to reduce the number of multihop relation paths. Lan et al [11] proposed modifying the staged query graph generation method to deal with longer relation paths and large search spaces.…”

Advancements in Complex Knowledge Graph Question Answering: A Survey

“…Then, a memory network [38] is employed to generate the final answer. Recent works [39,40] have introduced attention mechanisms and multi-column modules to improve the ranking accuracy in this framework. We summarize the procedure of GNN-based methods into four modules, as illustrated in Figure 4.…”