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

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 8805–8812 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 8805 Expanding End-to-End Question Answering on Differentiable Knowledge Graphs with Intersection Priyanka Sen Amazon Alexa AI Cambridge, UK sepriyan@amazon.com Amir Saffari Amazon Alexa AI Cambridge, UK amsafari@amazon.com Armin Oliya Amazon Alexa AI Cambridge, UK aooliya@amazon.com Abstract End-to-end question answering using a differ- entiable knowledge graph is a promising tech- nique that requires only weak supervision, pro- duces interpretable results, and is fully dif- ferentiable. Previous implementations of this technique (Cohen et al., 2020) have focused on single-entity questions using a relation fol- lowing operation. In this paper, we propose a model that explicitly handles multiple-entity questions by implementing a new intersec- tion operation, which identifies the shared el- ements between two sets of entities. We find that introducing intersection improves perfor- mance 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 perfor- mance on questions with multiple entities by over 14% on WebQuestionsSP and by 19% on ComplexWebQuestions. 1 Introduction Knowledge graphs (KGs) are data structures that store facts in the form of relations between enti- ties. Knowledge Graph-based Question Answering (KGQA) is the task of learning to answer questions by traversing facts in a knowledge graph. Tradi- tional approaches to KGQA use semantic parsing to parse natural language to a logical query, such as SQL. Annotating these queries, however, can be expensive and require experts familiar with the query language and KG ontology. End-to-end question answering (E2EQA) mod- els overcome this annotation bottleneck by requir- ing only weak supervision from question-answer pairs. These models learn to predict paths in a knowledge graph using only the answer as the train- ing signal. In order to train an E2EQA model in a fully differentiable way, Cohen et al. (2020) pro- posed differentiable knowledge graphs as a way to represent KGs as tensors and queries as differen- tiable mathematical operations. Figure 1: To answer “Who did Natalie Portman play in Star Wars Episode II?”, we identify all the characters Natalie Portman has played, all the characters in Star Wars Episode II, and intersect the two resulting sets to get to the answer, Padmé Amidala. Previous implementations of E2EQA models us- ing differentiable knowledge graphs (Cohen et al., 2020) have focused on single-entity questions us- ing a relation following operation. For example, to answer “Where was Natalie Portman born?”, the model could predict a path starting at the Natalie Portman entity and following a place of birth rela- tion to the correct answer. While this follow operation handles many ques- tions, it often struggles on questions with multi- ple entities. For example, to answer “Who did Natalie Portman play in Star Wars Episode II?”, it is not enough to identify all the characters Natalie Portman has played, nor all the characters in Star Wars Episode II. Instead, the model needs to find what character Natalie Portman has played …

Content truncated to 500 words