Research On Relation Extraction Method Based On Recurrent Convolutional Neural Network

As one of the important tasks in the field of natural language processing,entity relation extraction has been a research hotspot in academia and industry in recent years.The extraction of semantic relationships between entities has important implications for frontier domains such as information retrieval,automated question and answer,and intelligent recommendation.The traditional method of entity relation extraction requires manual design features,which consumes a lot of time and manpower,and can easily lead to error propagation problems.The neural network based method can automatically perform feature learning and avoid a lot of manual annotation work.Among them,the convolutional neural network has been gradually used in entity relation extraction tasks because of its excellent feature extraction capability.However,for a long entity co-occurrence sentence in a text corpora,it can only acquire local features and cannot learn long-distance dependence information.In this paper,the entity relation extraction method using a combination of a recurrent neural network and a convolutional neural network is studied.The main tasks are as follows:1.For the problem that simple convolutional neural network cannot learn long-distance dependence information,this paper proposes combining a recurrent neural network that is good at processing sequence information and a convolutional neural network that is good at extracting features,adding a recurrent neural network GRU that is good at handling long-distance dependencies to the vector representation stage of convolutional neural network,the contextual information vector of words is obtained through bidirectional GRU learning,which provides more abundant features for the following relation extraction model training.2.For the problem that the common max pooling in a convolutional neural network cannot capture the structure information between two entities,this paper proposes to adopt the piecewise max pooling method in the pooling layer of the convolutional neural network to extract the structure information between entities while extracting more fine-grained feature information.3.For the problem that incorrect annotation caused by distant supervised methods,this paper proposes to add a sentence-level attention mechanism to the relation extraction model,so that the sentences that express the corresponding relation correctly get higher attention and those who are annotated wrongly get lower attention weight,so as to weaken the interference of the wrongly annotated corpus to the model,and improve the accuracy of the entity relation extraction.This paper designs the experimental verification of English and Chinese corpus,and compares with the traditional methods and other methods based on neural network.It is proved that the method can effectively improve the precision and recall rate of entity relation extraction.