Judicial Case Recommendation System Based On Deep Learning

Due to huge the system of judicatory domain is and the complex of its content,there are a large number of judicial cases and related documents,making it extremely difficult to obtain satisfied information from massive judicial documents.To solve this problem,this thesis improves the content-based recommendation model by using deep learning and knowledge graph techniques.This model can recommend cases that satisfy users’ requirements well.First of all,the text classification model OCCNN based on character level convolutional neural network was proposed in this thesis to solve the problem of file classification of massive cases.The model can accurately and effectively solve the text classification problem by extracting feature through convolutional neural network.Experimental results show that the model can achieve 99.67% accuracy of classification,and the training time is only 50% of recurrent neural network.Secondly,this thesis proposed a semantic similarity calculation model TF-W2 V based on word frequency and word vector,which solved the semantic similarity calculation problem of judicial cases.The TF-W2 V model utlizes Word2 Vec to train the texts and get the word vectors.At the same time,TF-IDF algorithm is used to obtain the word weight.The feature vectors of the texts are obtained after cascading and compression,and thus the similarity of the text is obtained by cosine similarity algorithm.Experimental results showed that the accuracy of the model was improved by2% to 36% compared with other baseline models.Using DCG as the evaluation index,the maximum increase was 16.68 and the minimum increase was 4.Compared with conventional methods,our model can understand text semantics better and get more accurate semantic similarity of texts.Thirdly,this thesis constructs the knowledge graph of judicial cases.The thesis utilizes the LTP language platform to extracted knowledge triples from the judgment materials to construct the cases knowledge graph.Finally,this thesis combines the above text classification model,semantic similarity calculation model and judicial domain knowledge graph,to improve the content-based recommendation model,and makes the recommendation model more intelligent and able to understand the semantics and recommend cases to users.Finally,we used flask framework to implement the case recommendation model and obtain the case recommendation system based on deep learning.