Research on knowledge base question and answer methods based on the joint subgraph structure of interrogative features

Existing knowledge base question and answer methods rarely consider the influence of question sentences and candidate subgraph structure information when performing entity representation. This paper proposes a knowledge base question and answer method based on subgraph structure fusion question perception to address these issues. Firstly, the question is transformed into a semantic quantity by the pre-training model RoBERTa; secondly, the subgraphs of the candidate answer sets are extracted from the knowledge base based on the subject entities in the question, and the structural representation of the candidate subgraphs is enhanced by the attention calculation based on the query information to achieve question perception; secondly, the entity representation is pre-trained based on the background knowledge base and fused with the corresponding structural representation. Finally, the candidate answers are scored based on the fused vectors, and the entity with the highest score is taken as the answer. Comparative tests were conducted on the WebQuestionsSP dataset, and the experimental results showed that the proposed model outperformed other benchmark models.