Expanding End-to-End Question Answering on Differentiable Knowledge Graphs with Intersection

TL;DR

This paper enhances end-to-end question answering on differentiable knowledge graphs by introducing an intersection operation, significantly improving accuracy on multi-entity questions across two datasets.

Contribution

It proposes a novel intersection operation for differentiable knowledge graph question answering, enabling better handling of multi-entity questions and improving performance.

Findings

Performance improved on WebQuestionsSP from 69.6% to 73.3% Hits@1.

Performance improved on ComplexWebQuestions from 39.8% to 48.7% Hits@1.

Notable gains in multi-entity question accuracy, over 14% on WebQuestionsSP and 19% on ComplexWebQuestions.

Abstract

End-to-end question answering using a differentiable knowledge graph is a promising technique that requires only weak supervision, produces interpretable results, and is fully differentiable. Previous implementations of this technique (Cohen et al., 2020) have focused on single-entity questions using a relation following operation. In this paper, we propose a model that explicitly handles multiple-entity questions by implementing a new intersection operation, which identifies the shared elements between two sets of entities. We find that introducing intersection improves performance over a baseline model on two datasets, WebQuestionsSP (69.6% to 73.3% Hits@1) and ComplexWebQuestions (39.8% to 48.7% Hits@1), and in particular, improves performance on questions with multiple entities by over 14% on WebQuestionsSP and by 19% on ComplexWebQuestions.