Research On Intelligent Statutes Recommendation Technology Based On Collaborative Filtering And Text Relevance

In recent years,artificial intelligence technology has developed vigorously and is slowly changing people's production and lifestyle.China is committed to achieving high-quality development,improving the intelligence level of economic and social development enhancing public services and urban management capabilities through artificial intelligence technology.At the same time,the Supreme People's Court is also energetically promoting the"Intelligent Court",making full use of modern science and technologies such as Internet,Cloud Computing,Big Data,Artificial Intelligence and so on,to promote the modernization level of the trial system and trial capacity.Statutes recommendation is an important part of intelligent trial,because it is the basis of case trial results.If we can predict the statutes correctly,we can know the trend of the trial to some extent.In addition,the various roles involved in legal cases can benefit from the statutes recommendation.It can help judges deal with cases more efficiently;it can also help lawyers find stronger arguments to defend better;for people who lack legal knowledge,it is difficult for them to know whether there are appropriate statutes to protect their rights without the help of professionals,while cons ulting from law firms requires a lot of time and money.A system that predicts the statues correctly can solve their problemsBased on the analysis of the relationship between case information and statutes,this thesis proposes an intelligent statutes recommendation technology which combines collaborative filtering and text relevance.To be used by people without legal knowledge this thesis chooses the litigant claim of the case described by the natural language as input,and applies the statutes of the case as output.This method mainly consists of six steps which are the construction of the statutes database,the import of the cases database,the query of similar cases,the acquisition of alternative statutes set,the measurement of text relevance and the addition of accompanying references.The construction of statutes database includes importing common statutes from the database of Sichuan University and extracting remote statutes from judgment documents.The import of cases database refers to extracting effective information from the judgment documents,such as case number,litigant's claim and statutes quotation,discarding judgment documents missing the key information,preprocessing the litigant's claim,standardizing the statute name,and associating with the statutes information of statutes database.Querying similar cases is a collaborative filtering method,obtaining the main features of a case by vectorizing the litigant's claim through LDA topic model at first,and then calculating the Euclidean distance between the features,getting the most m similar historical cases by Ball Tree algorithm.The acquisition of alternative statutes set is to calculate the scores of the quoted statutes of similar historical cases and reorder,selecting the first n items as alternative statutes set,so as to reduce the scope of statutes by collaborative filtering.Text relevance measurement refers to matching the relationship of the litigant's claim and the alternative statutes one by one through a novel CNN model to obtain the text relevance score,and then putting the score into the logistic regression model for prediction together with statute ranking.If the prediction is relevant,the statute will be recommended;if not,it will be discarded directly.Adding accompanying references refers to using the Apriori algorithm to find the association rules of statutes and extending the existing recommendation results.This method reduces the scope of alternative statutes by collaborative filtering,and solves the problem that there are too many statutes to train.This method also has the advantage of strong scalability,which can be applied to new published statutes without changing the model architecture.This method also adds the statutes content as input,which effectively improves the accuracy of the statutes recommendation by text relevance matching.Moreover,all kinds of features used in this method are automatically extracted by machines without human assistance.Different from the traditional multi-classification method,this method does not have the problem of category bias caused by uneven distribution of samples,and can recommend statutes of different number for different cases,instead of fixing the top K.Finally,this thesis has carried out a series of experiments,compared with other text recommendation methods,proved this model is suitable to intelligent statutes recommendation,and analyzed the advantages and disadvantages of these methods.In the end,this thesis described the direction of further improvement.