Entity Linking Task Of Chinese Short Text

At present,with the continuous development of mobile information platform,the demand for Internet natural language processing is also increasing.In recent years,with the increasing popularity of Weibo,We Chat,QQ and other short text information platforms,short text information is growing exponentially.Short text information contains huge data value and broad market prospects.How to dig out this part of the value has become a topic worth discussing.It is an important part of the field of short text language processing to build an entity chain between text entity allegations and knowledge.Focusing on improving the effect of Chinese short text entity chain pointing,this paper has done the following work on the basis of the existing research:(1)Based on the Bi LSTM-CRF neural network model,the self-attention mechanism is introduced to construct the Self Attention-BILSTM-CRF neural network model for the named entity recognition task of Chinese short texts.By integrating the self-attention mechanism,the model can automatically capture the internal information of the text,understand the semantic meaning of the text better,and dig the sentence features without the help of external data.This method can automatically realize named entity recognition without manual annotation.Through comparative experiments,it is proved that this model has a better recognition effect.(2)Based on the ERENIE pre-trained language model proposed by Baidu,through adding multi-task learning this paper builds a Chinese short text entity disambiguation model.By using the entity classification task as the auxiliary task of the Chinese short text entity chain finger,multi-task learning is carried out,which improves the problem that the entity mention and candidate entity correlation calculation basis is insufficient caused by the insufficient context information in the process of the Chinese short text entity chain index,and improves the effect of the solid chain finger of the model and enhances the generalization ability of the model.Through comparative experiments,it is verified that the effect of the model based on multi-task learning is better than the single task model with the same structure,and it is proved that the model can effectively solve the Chinese short text entity chain pointing task.