An Effective Simple Question Answering Framework using Knowledge Graph Embedding

This is the authors' abstract. We don't add key points for this paper.

Question answering over knowledge graph(KGQA) is an important research direction in question answering(QA) tasks. It is given a natural language question, which is answered by semantic understanding and parsing of the question using facts from the knowledge graph(KG).Currently, the main challenges for simple KGQA are the ambiguity of entities and the accuracy of relation linking. To bridge this gap, we propose an effective KG embedding based simple question answering framework (KGESQA) that embeds entity and relation in the KG into a continuous vector space, thus preserving the structural information in the KG while facilitating computation. We then used KG embedding to improve the effectiveness of entity linking and relation linking. The method uses entity type information to entity disambiguation, then embeds the subject entities of questions into the KG embedding space to find their corresponding entity, and uses the textual information of the word embedding and the structural information of the KG embedding to improve the accuracy of relation linking. Experiments conducted in a widely adopted benchmark test show that our proposed KGESQA outperforms state-of-the-art methods.