A relation extraction method and system based on sentence entity relation network

Abstract

The present invention relates to a method and system for extracting a relationship based on a sentence entity relationship network, and the method comprises: constructing a sentence entity relationship network according to a training data set; expanding the sentence entity relationship network according to text data triples of document data to be extracted; querying candidate sentences having common entities with a target sentence from the sentence entity relationship network, and calculating the similarity between each candidate sentence and the target sentence using a Sentence-Bert model, and taking the candidate sentences with similarities greater than a preset threshold as similar sentences; extracting key words related to the common entity from similar sentences using an attention mechanism; and preprocessing key words and text data triples using a Bert model and then feeding them into a classifier for relationship extraction to obtain semantic relationships between entities. The present invention can not only obtain richer information about entities in sentences to be extracted, but also avoids heterogeneous information fusion, thereby improving the accuracy of information extraction.

Images