Transition-based Neural Network With Joint Multi-task Label Space For NER

Named Entity Recognition(NER)is one of the fundamental tasks in the field of Natu-ral Language Processing.It is an important foundation for Information Extraction,Machine Translation,Question Answering Systems and other fields.Named Entity Recognition can extract person names,place names,organization names from texts and help to store texts.The fundamental tasks in the field of Natural Language Processing also include Chinese Word Segmentation,Part-of-Speech Tagging and Constituent Parisng.These tasks are usu-ally performed independently and the results of these tasks may be inconsistent.Actually,by jointing Named Entity Recognition with Constituent Parisng,the latter can help to improve the effectiveness of the former.Accordingly,the main contents of this paper are as follows:(1)A Chinese Named Entity Recognition model with joint multi-task is proposed.The improvement of Named Entity Recognition by jointing Constituent Parsing can be achieved with this model.For a sentence,the model can output the analysis results of the four tasks mentioned above;(2)The model calculates the embeddings of different features from the analytical states with LSTM and Bi-LSTM when combines Chinese Named Entity Recognition with other tasks.These embeddings can be concatenated to form an embedding and different tasks share this embedding in the model.The experiments of this paper use the datas from Onto Notes 5.0.Experiments are car-ried out on whether the Constituent Parsing is included in the joint model and the results of the experiments show that Named Entity Recognition can be improved by Constituent Parsing.Jointing Constituent Parsing,the F₁score of ORG improved 4.77%.Moreover,through different comparison experiments,the effect of different model parameters on the performance of the model was verified.