Research On Entity Relation Extraction Method Based On Deep Learning

With the advent of the era of digital,information,network information to exponential growth speed,make people receive far more than they can bear or need information,of the growing phenomenon of multidisciplinary cross fusion,how accurate and efficient derives from the vast data information on human society valuable and meaningful information,become the era of big data needs to be an important problem in seeking the way of crack.As the basis of knowledge graph construction,entity relation extraction task aims to judge the semantic relations between entity pairs in unstructured text and divide them into preset types,so as to make the data into structured form.Therefore,it has important research value.At present,most of the entity relation extraction models based on deep learning have problems such as error propagation,insufficient extraction of deep features from corpus and insufficient utilization of polysemy information.In view of the above problems,this paper conducts research from the following two perspectives:1.According to the noise emitted by distant supervision and annotation data and deep learning problem of error propagation in feature extraction,this paper,by using residual network can be the characteristics of shallow and deep information jointly said to establish deep convolution neural network,and by changing the identity mapping in the transfer function,in order to achieve the deep constantly update feature weights in the transfer process,access to a deeper and more effective in the text of the input information,has reached the higher classification accuracy.2.As the existing entity relation extraction methods do not consider the co-occurrence relationship between entity pairs and language structure information,this paper combines the pre-training language model with multiple semantic information to construct a multisemantic fusion entity relationship extraction model.The pre-training language model is applied to the entity relation extraction task,and the semantic vector and language structure vector are combined to highlight the semantic feature information,which can alleviate the problem of insufficient representation of the local and global information of the sentence,and improve the utilization of the local and global features of the text.Further,this paper considers the semantic differences between the co-occurrence and expression of different entities in the sentence,and introduces the co-occurrence information of entities as the semantic feature to solve the polysemy problem.This paper combines entity co-occurrence information with pre-training language model to construct entity relationship extraction model,and conducts experimental verification on Chinese and English datasets respectively.Experimental results show that the model proposed in this paper can improve the extraction effect of entity relation more effectively.