Recommendation Of Legal Articles Based On Deep Learning

With the development of artificial intelligence,judicial intelligence has entered a process.The ever-increasing number of cases makes it difficult for professional lawyers to get involved in all the legal provisions,and it is time-consuming and labor-intensive to find appropriate legal provisions to support this case from a large number of legal provisions.Therefore,how to use artificial intelligence to promote judicial informatization,improve the efficiency of caseworkers and protect the judicial needs of the people is an urgent problem.Based on the previous research work such as deep learning and text classification,this paper studies the article recommendation problem,which is mainly divided into four steps.First,build a legal corpus,preprocess the data and use Jieba Chinese word segmentation,and then use TF-IDF to select keywords and put them into the dictionary.Second,use Word2 vec to train the word vectors on the extracted legal documents to form a word vector model.Third,a single deep learning model is used for multi-label classification tasks,namely deep convolutional neural network(CNN)and attention model(ATTENTION).Fourthly,in view of the shortcomings of single model prediction,this paper adopts the method of model fusion to improve the convolutional neural network fusion model(ATTCNN)that integrates the attention mechanism,and proves that the fusion model can indeed achieve better laws Recommended effect.The convolutional neural network model has a better performance in extracting text features.The attention model is a neural network that has recently developed rapidly in the field of natural language processing,and it improves the problem that the performance of the recurrent neural network will decrease as the length increases.Based on this background,this article uses the CNN and ATTENTION models in conjunction with the relationship between the case description and the applicable legal provisions.The accuracy rate of the CNN model reaches 97.6%,the micro average is 37%,and the macro average is 21.4%.Although the CNN accuracy rate is very high,it is not enough to look at the classification accuracy rate for extremely skewed data,and this article is a multi-classification task.Macro average and micro average display are not good.The accuracy of the ATTENTION model is 89.7%,the micro average is 69.2%,and the macro average is 53.6%.The micro average and macro average of the attention model have improved significantly.However,the two single models are not particularly effective.In order to fuse a single model.Convolutional neural networks are weak in extracting the features of legal text data and cannot reflect the relationship between context and keywords.The ATTENTION mechanism has a good effect when learning long-range dependent sequence information.Use the ATTENTION mechanism to extract the semantic relationships of the case descriptions and fuse the convolutional neural network to obtain a new network model ATT-CNN.The model accuracy rate is 95.6%,the micro average is 80.9%,and the macro average is 77.4%.The comparative analysis shows that the prediction effect of the fusion model has been significantly improved,and to a certain extent,it can be used as an auxiliary method recommended method.Through the above research work,this article puts forward a feasible research method and theoretical basis for the relevant legal recommendations.