Research On Mongolian Knowledge Graph Automatic Q&A For Legal Field

With the gradual improvement of China’s legal system,the demand for legal consultation from the masses is increasing.Nowadays,the Mongolian people still stay in the way of manual consultation,which leads to the loss of balance between the amount of legal consultation and the resources of lawyers,and the human cost of legal practitioners becomes higher.Mongolian automatic question and answer system in the legal field can effectively alleviate the shortage of manual legal consultation.Mongolian legal question and answer system is facing difficulties,such as strong professionalism,few employees and few data resources.This thesis studies the automatic question and answer technology of knowledge graph for legal field,and designs and implements an automatic question and answer system of Mongolian knowledge graph for legal field,aiming at the shortcomings of the existing legal question and answer database limitation,inability to reason and low specialty in reply.The main research contents are:(1)Construction of Mongolian Data Resources for Automatic Question and Answer System in Legal FieldThis thesis first collects,screens,translates,manually corrects,and automatically corrects existing Chinese legal data resources,and constructs a corpus of 150,000 pairs of Mongolian legal questions and answers.Then,from the Mongolian question and answer corpus,the entity extraction annotation corpus of 17400 sentences and the attribute annotation corpus of16200 sentences were constructed to provide data support for the semantic analysis of the question sentence.Finally,in view of the current situation of the lack of knowledge graph in Mongolian legal field,we constructed the legal knowledge graph of 857 entities related to criminal offences.(2)Research on Mongolian Entity Extraction in Legal FieldFirst,BiLSTM-CRF model is used to study the Mongolian legal entity automatic extraction method.To increase the generalization of the model,the multilingual pre-training model CINO is selected and the entity extraction model based on CINO-BiLSTM-CRF is studied.In addition,in order to make use of the validity of the Transformer model and BiLSTM model in extracting context long-distance dependence and directional features,a Mongolian entity extraction model based on Transformer-BiLSTM-CRF is studied by combining the Transformer model and BiLSTM model for feature extraction.The results show that the introduction of pre-training CINO model can effectively improve the generalization ability of entities with less training data in BiLSTM-CRF model,and CINO-BiLSTM-CRF model improves the values of P,R and F1.By combining the feature extraction capabilities of Transformer model and BiLSTM model,and capturing both long-distance and directional features.Therefore,Transformer-BiLSTM-CRF model achieves the best comprehensive effect,with the F1 value of 93.30% and the performance on nested entities is also the best.(3)Research on Mongolian Entity Attribute Extraction in Legal FieldBased on the multilingual minority pre-training CINO model,we designed two Mongolian attribute extraction models,such as CINO-BiLSTMAttention and CINO-BiGRU-Attention model.The results show that using the BiLSTM model for feature extraction on Mongolian sentences is better than the BiGRU model,and the CINO-BiLSTM-Attention model outperforms the CINO-BiGRU-Attention model in P,R and F1 metrics.(4)Mongolian Intelligent Question and Answer System for Legal DomainIn view of the disadvantage that the results of retrieval legal question and answer are limited by the corpus,this thesis improves the accuracy of question and answer by using Mongolian legal knowledge graph auxiliary retrieval.The system is mainly divided into question semantic analysis module,knowledge graph retrieval module,and retrieval automatic question answering module.Firstly,the Mongolian legal entity extraction model and attribute extraction model are used to analyze the Mongolian question semantics.Then,the knowledge graph question answering system and the retrieval question answering system are integrated to improve the response accuracy of the automatic question answering system for the legal field,and the accuracy rate of the retrieval question answering system is increased by2.13%.