Text-Enhanced Reasoning for Question Answering over Incomplete Knowledge Base

The construction and maintenance of a structured Knowledge Base is too time-consuming and laborious to cover all the world’s knowledge. When facing some questions, the Knowledge Base will lack its key evidence information or answers, resulting in ineffective reasoning of the Question Answering system. A prevalent solution to make up for an incomplete Knowledge Base is the utilization of Text Corpus. However, previous methods fail to thoroughly mine the topological structure information in Knowledge Bases, and their limited interaction between the two heterogeneous knowledge hinder the effective use of text knowledge in answering reasoning. To solve this problem, we propose a Text-Enhanced Reasoning Model for Question Answering over Incomplete Knowledge Base. It focuses on the topological structure present in the Knowledge Base, as well as the clues derived from the text corpus, and emphasizes the interaction and fusion of information between these two sources. We simulate incomplete scenarios with 10%, 30%, 50% and 100% Knowledge Base and conduct extensive experiments on an open dataset WEBQUESTIONSSP. The results surpasses the current state-of-the-art methods, demonstrating that our model effectively utilizes text information to enhance the reasoning ability of Question Answering over Incomplete Knowledge Base.