A Dataset and Multi-task Multi-view Approach for Question-Answering with the Dual Perspectives of Text and Knowledge

Abstract

Question-answering (QA) systems are important tools for extracting information from large datasets and providing accurate and relevant answers to user queries. Two of the most widely studied and built QA systems are Natural Language Question Answering (NLQA) and Knowledge Graph Question Answering (KGQA). NLQA relies on sequence learning algorithms, which have limitations on the length of input they can handle, while KGQA relies on the Subject-Predicate-Object (SPO) tuple representation of data, which may not always be available in the knowledge graph. In this paper, we present a novel approach for addressing these challenges by utilizing the structural information from the Knowledge Graph (KG) and the semantic information from the Natural Language Context. Due to the lack of a dataset to enable this approach, we propose the creation of a multi-view dataset - MTL-QA, specifically designed for multi-task learning. We also present a multi-task learning approach to jointly train NLQA and KGQA models and demonstrate the effectiveness on the proposed MTL-QA dataset.