Reasearch On Entity Relation Extraction In The Field Of Party Building

With the explosive growth of information on the network,more and more information extraction technologies are needed to process text automatically..Relation extraction is a subtask of information extraction.Its main goal is to mine the relationship between entities.At present,there are many studies on the relation extraction of biology,medicine or personal relationship categories in Chinese corpus.However,with the rise of "Internet + Party Building",relation extraction for Party Building data has become a task that needs attention.It is of great significance to the construction of knowledge map,question-and-answer system in the field of Party Building and to the promotion of the process of Wise Party Building in China.There are two commonly used methods about relation extraction.One is the feature vector based methods,which relies on artificial construction features.Then the features are transformed into vectors and input into a suitable machine learning algorithm.The performance of the algorithm will be affected by the quality of the features.The other is kernel based method,which are not suitable for large-scale data sets.It uses neural networks to automatically learn features,reducing the dependence on labor.Therefore,how to use the deep learning model to solve the problem of the relation extraction in the Party Building field is the research focus of this paper.As for the dataset in the field of Party Building,this paper proposes MA_BGRU,which use BGRU to model sentences and introduce Multi-head Attention,aiming at getting more information about sentences from different representation spaces and improving the expressive ability of the model.Meanwhile,based on the existing word and position feature as network input,the dependent syntax feature and the relative core predicate dependency feature are further introduced.The dependency syntactic feature includes the dependency value of the current word and the location of the dependent parent node,so that the model can further obtain more text syntactic information.The experimental results on Party Building data show that the model proposed is very effective for the task of entity relation extraction in the field of Party Building.