Research on Chinese Relation Extraction Based on Multi-level Semantic Perception
Relation extraction is the basis of constructing knowledge graphs,and Chinese relation extraction is also a difficult problem in relation extraction.Most existing Chinese relation extraction methods use character-based or word-based features,but the former cannot capture contextual information of characters and the latter is limited by the quality of word segmentation,resulting in lower performance of Chinese relation extraction.In response to this problem,a Chinese relation extraction model based on multi-level semantic perception is proposed.This model uses rich semantic information between entities to improve the performance of predicting relationships between entities.Multi-level semantic perception is reflected in the following three aspects:firstly,text information is transformed into dynamic word vectors using the pre-training language model ERNIE;then,attention mechanism is used to enhance the semantic representation of the sentence where the entity is located,while external knowledge is used to eliminate Chinese ambiguity of entity words as much as possible;finally,the sentence representation containing multi-level semantic perception is put into classification for prediction.Experimental results show that the proposed model outperforms existing models in Chinese relation extraction performance and is more in-terpretable.
国家科技期刊平台
NSTL
万方数据
维普
