An entity relation extraction method based on the fusion of contextual information

Abstract

With the growth of information technology, numerous entity relation extraction methods have developed. However, current studies primarily emphasize effective entity recognition, disregarding the significance of local information about entities in text on relation extraction and only acknowledging the presence of a singular relationship. To address this, a proposed entity relation extraction method based on the fusion of contextual information integrates Bi-RNN to further extract sentence vectors encoded by BERT.The model combining local information about entities with context, facilitating multi-relation classification through a biaffine classifier. Additionally, the model reduces the dimensionality of fused information to capture more effective information. Negative sampling is also introduced to enhance generalization capabilities. The model outperforms existing works in multiple public datasets.