Research On Chinese Entity Relationship Extraction Based On Deep Learning

With the development of Internet technology.there is an explosive growth of information.The formats of the information are mostly unstructured data which is largely made up of text data.Therefore.how to convert the unstructured information mentioned in this paper into structured information that helps with computer's comprehension has become a critical problem of data mining,and entity relation extraction is the research hotspot of it.The extraction method of traditional pipe model is to regard entity identification and relation extraction as two individual and separate subtasks,however,a lack of relevant information between entity and relation leads to some mismatches in the model.Hence,researchers tried the joint extraction.Though the previous joint extraction model improved the problem of error propagation in a certain extent,However,due to the limitations of the traditional fixed expression method of word vectors and the extraction capability of feature extractors,it is still difficult to break through the bottleneck in improving the overall effect of the model.Based on previous research.A BERT-based pre-training model architecture proposed in this paper is used in extraction of entity and relation task,and the pre-trained relation extraction model is fused with named entity characteristics,speech tagging characteristics.And the structure of the Bidirectional Long-Short Term Memory Neural Networks and the Bidirectional Gated Recurrent Unit Neural Networks are added to enhances the feature extraction ability of the model.Pre-training was carried out on Baidu Baike data set with larger corpus in advance.Thus to further improve the effect of model extraction.In addition,a new extraction program is adopted,which is to firstly extract the text body in recognition stage of main entity,and then running a global prediction for relation and the object that probably connected with main body in object recognition stage.By comparing different models on SKE Chinese data set which is known for the high-quality tagging and the largest scale and operating ablation experiments on the model,the model isverified to be advanced and effective.