| Community correction is an important system in China,which is beneficial to optimize the allocation of penalty resources and improve the quality of education reform.Whether the community correction work can achieve good results largely depends on whether the correction scheme formulated by the community correction staff is reasonable.With the comprehensive development of community correction in the country,the number of community inmates has increased significantly.At the same time,due to the improvement of the community correction system,the information of community inmates has become more and more abundant,which makes it more difficult for the correction staff to formulate the correction scheme.At present,the formulation of the correction scheme completely depends on the comprehensive analysis of the information of community inmates,which needs a lot of time and energy.However,at this stage,the correction staff equipped by the judiciary is limited,the task is heavy,and the situation of community inmates can not be deeply understood.As a result,the correction scheme is mere formality and the implementation effect is poor.On the other hand,the formulation of the correction scheme depends heavily on the experience of the correction staff,which is subjective.The lack of theoretical and practical experience of the correction staff may lead to the uniformity and lack of pertinence of the correction plan.In view of the above problems,this thesis puts forward a feasible method to recommend community correction scheme,and the main work is as follows:Firstly,a correction scheme recommendation method based on text similarity and association rules is proposed.Through the analysis of typical cases of community correction,the content is divided into two parts.In the first part,the basic information belongs to short text,which is vectorized by word2vec based on information entropy and inverse document frequency weighting.In the second part,the information of crime and living environment belongs to long text,which is modeled by doc2vec after specific text preprocessing.According to the different contribution of the two parts to the overall information,a similarity measurement method based on weight allocation is proposed.The similarity of the two parts is calculated and given different weights.Then,based on collaborative filtering,the correction scheme accepted by the community inmates with the highest similarity is recommended.Through the analysis of the correction scheme,it is found that some of the correction measures often appear together,and there is a certain association relationship.This paper proposes a method of mining association rules from frequent itemsets.By analy zing the rules of the collaborative occurrence of the correction measures,valuable association rules are mined to supplement the existing recommendation results.Secondly,a risk assessment method for community inmates is proposed.Aiming at the practical problems of recidivism data set,such as more sensitive information and more features,the data set is preprocessed and feature selected.In order to solve the imbalance problem of recidivism data set,the smote algorithm is used for over sampling,and then the data cleaning technology Tomek links is introduced to delete the noise points,and then the risk assessment model is established based on xgboost algorithm.According to the risk assessment results,more strict supervision and management corrective measures should be added to the community inmates with higher risk.Finally,on the basis of the above research,through the analysis of the actual needs of community correction work,the community correction scheme recommendation system is designed and implemented.The system can recommend the correction scheme for the designated community inmates,and support the correction staff to review and modify the given scheme. |
