Research And System Realization Of Recognition And Normalization Of Scientific Research Entities

With the increasingly prominent role of science and technology in social development,the analysis of scientific research data becomes more and more important.Therefore,this subject focuses on the information extraction of scientific research data.Named Entity Recognition is a very important part of many natural language processing tasks,but the accuracy has not yet reached our expectation,especially in the task of Chinese domain.The focus of this subject is named entity recognition of Chinese research data and its normalization.First,this thesis presents a network structure for entity identification of scientific research data that combines multi-size convolution kernel with LSTM,which can effectively enhance the recognition effect.In the meantime,it has been observed that the recall rate of some other entity recognition algorithms is much lower than the accuracy,so this subject also attempted to alleviate this situation by changing the loss function.Experimental results show that the new loss function can not only solve the situation of the recall rate being too low,but also greatly raise the training speed.Second,due to the reduntant representations of a single scientific entity with identified intrinsic meaning,we study the entity normalization algorithm based on the Siamese network structure.With a series of experiments and the final application of the system,this subject proves the correctness and effectiveness of the proposed algorithm and system design.