Research On Relation Extraction Method Based On Entity Boundar

The relation extraction aims to extract semantic relationships between entities from text.In the traditional pipeline model,named entity recognition and relation extraction are regarded as two independent sub-tasks respectively.The named entity recognition is performed first,and then relation extraction is performed.Because relationship extraction is the upper task of named entity recognition,errors in named entity recognition will lead to error diffusion,and the dependency information between the two sub-tasks can not be fully utilized,and finally affect the overall performance of named entity recognition and entity relation extraction.In order to effectively alleviate and solve the problem of error diffusion and neglect the relationship dependence between the two sub-tasks,this paper proposes a relationship extraction method based on entity boundary,as follows:(1)Aiming at the problem of error diffusion,this paper proposes a relation extraction method based on entity boundary combination.Through the research of a nested named entity recognition model based on deep boundary combination,it is found that the F1 value of entity start boundary is 94.06,and the F1 value of entity end boundary is 94.88,and the final entity recognition performance is 80.12,which is lower than that of entity boundary recognition.Compared with entity identification,entity boundary has small granularity and ambiguity,is easier to identify,has higher accuracy and performance,and can effectively alleviate the problem of error diffusion.Therefore,this method does not directly use entities to extract relationships,but skips entities and adopts entity boundaries to extract relationships.Moreover,the entity type feature and location feature are added into the model by feature combination method,so that the model can further consider the structural information of sentences,so that the performance is further improved,and the impact of error diffusion is reduced again.The experimental results show that the F1 value on THE ACE2005 English dataset reaches 76.21%,which is 14.11 percentage points higher than the comparison method.(2)Aiming at the shortcomings of pipeline method,this paper proposes an end-to-end extraction method based on entity boundary regression relationship based on entity boundary combination.Different from the pipelining method,which separates the two subtasks,this method uses the boundary regression method to identify the position of the entities in the sentence and the relationship between the entities.This method is an end-to-end approach,which can effectively solve the problem of error diffusion and neglect of relationship dependence between two subtasks in pipeline method.At the same time,the application of this method to the boundary regression algorithm is a new attempt in relation extraction task,which has a positive role in promoting the future research of relation extraction.Experimental results show that the boundary regression model achieves good performance on ACE2005 English data set,which verifies the effectiveness of the boundary regression algorithm in relation extraction.