A Semantic Enhanced Chinese Entity Relation Extraction Model Based on Relation Constraints

As an important subtask of natural language processing, the purpose of entity relation extraction is to extract entity relation triples from unstructured text data. The existing cascade entity relation extraction model has achieved satisfactory results on English dataset. However, due to the differences in literal semantics and language patterns between Chinese and English, Chinese sentences contain more complex semantic information and there is a problem of nested entities, which leads to the existing cascade model's inability to accurately represent the deep semantic information of Chinese sentences, inaccurate relation classification and the problem of identifying redundant and erroneous entities. For this reason, this paper proposes a semantic enhanced Chinese entity relation extraction model based on relation constraints(SECRC). Starting from the complex Chinese semantics, this model embedding sentences based on the multi-granularity semantics of character, words and sentences, and realizes the comprehensive expression of sentence depth semantics; a relation attention entity extractor (RAEE) is proposed, which constrains the entity recognition process by extracting the relation information, thus improving the accuracy of Chinese entity recognition. The model was experimented on the three Chinese datasets of DuIE, FinRE and SanWen. The results show that this model successfully obtains the deep semantic expression of Chinese sentences, effectively improves the accuracy of the relation classification and entity recognition results of Chinese sentences, and is superior to other baseline models in effect.