Relationship Detection Model Based on BERT-BiLSTM and Multi-information Fusion
JI Jie
SHAO Qing
SHEN Ying
Abstract：Knowledge base question answering system(KBQA)has been widely used in tasks such as search engines,chat ro-bots,and graph retrieval.It answers questions through knowledge triples in the knowledge base,among which the task of relation de-tection is the core question.The traditional relationship detection model only uses the similarity between the question vector and the relationship vector to express the connection between the question and the relationship.There is a problem of single information,which results in a decrease in the accuracy of answering the question.In this paper,a BiLSTM model(FI-BERT-BiLSTM)based on the BERT pre-training structure and fusion of multiple information elements is proposed.First,a BERT pre-training layer is add-ed to the model to make up for the lack of information.Second,an information fusion layer is built to strengthen the problem and re-lationship characteristics,and perform secondary feature extraction on the data of public network layer.Third,It increases problems and diversity of relationship characteristics through the attention mechanism.Simulation experiments on two data sets,SimpleQues-tion and WebQuestion,the effectiveness of the model is verified.
Keywords：knowledge base question answerrelation detectionBERTdeep learningbi-directional attention mechanism
Publication Date：2024-09-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：8( 2626-2633 )