Research On Assisted Judgment Prediction Of Mongolian And Chinese Legal Documents Based On Deep Learning

With the development of society,the construction of legal system in China has been further strengthened.As the number of cases grows,it has brought many challenges to the legal practitioners.Whether fairness and justice can be maintained become a difficult problem.Under the background,legal judgment prediction arises at the historic moment.Legal judgment prediction is mainly to give a factual description of a case,then analyze the key information,and give a prediction result.In this way,not only to ensure that the judicial personnel to the same case with the direction of consideration,but also for the masses to provide an effective means of consultation for the intelligence of the judiciary and to ensure the justice of the law have provided an effective support.At present,legal judgment prediction is mainly concentrated in the field of Chinese and English.However,in terms of minority languages,there are relatively few researches on legal judgment prediction.For example,the research on legal judgment prediction for Mongolian is still to be explored.First,compared with the abundant corpus resources in Chinese and English,there is no public corpus in Mongolian.In addition,the Mongolian also has the problem of ambiguity,the traditional word vector representation method can not distinguish the meaning of words in different contexts.Based on the above problems,we builds a Mongolian corpus of a certain scale to study the legal judgment prediction based on Convolutional Neural Network(CNN)and Long Short Term Memory(LSTM)models.The main work is summarized as follows:(1)Automatic extraction of Mongolian and Chinese criminal judgment documents based on rules:corpus is the key to natural language processing,and its quality determines whether further research can be carried out.To conduct further research,we crawl data from China Judgements Online to solve the lack of Mongolian corpus.We combine with the existing system of criminal judgment documents tagging,extracting key information from Chinese and Mongolian data based on rules.Experimental results show that this method can effectively extract the key information from the judgment documents,and the F1 reaches 92.88%in Chinese and 93.14%in Mongolian.(2)Aiming at the problem that Chinese is prone to ambiguity and so on,we proposes to integrate the part of speech features into the coding end,and then train the word vector.Through the experiments in CNN and LSTM models,the integration of part of speech features can effectively alleviate the problem of Chinese legal judgment prediction and improve the effect of the model.To solve the problem of misjudgment in confusing cases in criminal cases,we proposed to combine the Attention mechanism into CNN and LSTM models.The experiment shows that the combination of attention mechanism can alleviate the misjudgment of easily confused cases,and further improve the effect of the model.(3)In view of the language characteristics of Mongolian,we have taken a specific treatment for it.With the diversification of current word vector training methods,we explored the influence of Mongolian word vector in different ways on the Mongolian legal judgment prediction task.And legal judgment prediction task sharing information between two task,we use the legal judge based on topology dependent model prediction tasks,contrast unused CNN model of sharing information and the use of shared information between two tasks LSTM model effect,thus to explore the various utilization of shared information between subtasks for legal judgment prediction effect.