Research On Named Entity Recognition For Chinese Legal Texts

In recent years,the application of artificial intelligence technology in the judicial field has received extensive attention from researchers.In particular,the intelligent analysis and processing to the mass legal texts has become an important content of judicial artificial intelligence research.It is crucial for other tasks such as syntax analysis and semantic analysis for legal texts to recognize named entities correctly from legal texts.However,there are few relevant literature about this reasearch.Therefore,the methods based on deep learning of named entity recognition for chinese legal texts are studied experimently.Specifically,the main work of this thesis are as follows:(1)A legal named entity recognition dataset consisting of legal case texts is developed.The work of developing the dataset includes analysing the legal texts?designing the suitable annotation and making the assistant tools.(2)The character-level neural network model for legal named entity recognition is implemented as the baseline system.The baseline system is a character annotation model based on the LSTM-CRF model by regarding the chinese named entity recognition task as a sequence labeling problem.The model uitilize context information by LSTM and then assigns label to every character by CRF.(3)The segment-level neural network model for legal named entity recognition is proposed.Chinese texts have no word delimiters,therefore,the chinese named entity recognition task can be divided into two sub-tasks:word segmentation and entity recognition.Obviously,it is more reasonable to assign labels to segments than to assign labels to characters.The GCNN-LSTM model for segment-level legal named entity recognition is proposed,which solves word segmentation and entity recognition by combining GCNN with LSTM based on beam-search algorithm.(4)The combination method based on two neural network named entity recognition models is proposed.The combination method builds a model combined the character-level neural network with the segment-level neural network.A character-level model is introduced when assigning a label to an entire segment,taking into account the segment-level feature and the character-level feature.The experimental results show that the segment-level named entity recognition method has better performance than the character-level named entity recognition method,and the combination method obtains better performance than the others.