关键词: 关系抽取, 预训练模型, 词嵌入, 特征融合, 语义理解

Abstract: Relation extraction task is aimed to extract the relationship between entity pairs from text, which is one of the hot directions in the field of natural language processing. In view of the problem that the grammar structure of the text in Chinese character relation extraction corpus is complex and the semantic features of the text can’t be learned effectively. A Chinese character relation extraction model based on pre-training and multi-level information was proposed for this reason. Firstly, the word vector was generated by the pre-training model which possesses powerful semantic representation ability. Then the original sentence was divided into sentence level, entity level and entity adjacent level for feature extraction. Finally, the relation prediction was performed by the information fusion of the sentence structure features, Entity meanings, and dependency relationship between entities and adjacency words. The experimental results on the Chinese people relationship dataset show that the model has an precision rate of 81.5%, a recall rate of 82.3%, and an F1 value of 81.9%, which is an improvement compared to the baseline model. Moreover, the F1 score on the SemEval2010-task8 English data set reaches 81.2%, which proves that the model has a certain generalization ability for English corpus.

Key words: relation extraction, pre-training model, word embedding, feature fusion, semantic understanding