Research On The Recommendation Method Of Community Correction Strategy Based On Reinforcement Learning

The community correction system is an important part of the national governance system.It not only conforms to the trend of internationalization,but is also a penalty enforcement system reform for the implementation of my country’s criminal policy of combining leniency and strictness.In recent years,with the continuous deepening of community corrections,the proportion of persons serving sentences of suspended sentence,parole,and temporary out-of-prison sentence has gradually increased.The need for rapid and accurate information analysis of corrections has increased the pressure on community corrections staff.How to realize automatic and accurate analysis of corrections data is very important.The current correction personnel information has the characteristics of large amount of data and strong concealment.The traditional model of community corrections uses judicial personnel to manually formulate correction strategies,and there are problems such as insufficient community corrections and inconsistent correction levels.In order to solve the problem of the differences in correction levels between different regions and cities,and to improve the quality of correction strategies,the requirements of using relevant technologies to realize the informatization and intelligence of community corrections are put forwardThe traditional community correction model adopts the method of artificially formulating correction strategies,which has problems such as strong subjectivity and inconsistent judicial level.Reinforcement learning has the characteristics of autonomous learning and is an extremely efficient way to identify data patterns.It can discover the connection between the correction strategy and the correction personnel information through the learning of the correction strategy.Due to the large amount of data for correcting personnel information,the method of manually formulating correction strategies is inefficient.The recommendation system can analyze a large amount of data,efficiently extract cross-features,and implement scientific correction strategy recommendations.Nowadays,the combination of reinforcement learning and recommendation system has been developed rapidly in the industry.In the process of recommending correction strategies,it is of great value to accurately portray the user’s portrait and psychological characteristics of correction personnel.How to find out how to integrate relevant information of correction personnel into the recommendation process under different circumstances to realize the personality of correction schemes of correction personnel Recommendations,and realize the cross-domain upgrade of judicial community work based on the research of crime impact factors.This paper constructs an accurate matching method based on the subspace clustering method for corrective programs,and on this basis,constructs accurate recommendations for corrective programs based on reinforcement learning,and provides ideas for improving the working methods of the judicial community.Aiming at the recommendation problem of the correction staffs personalized correction strategy,this paper studies a correction strategy recommendation algorithm that combines the correction staffs characteristic analysis based on the actual scene and combined with the characteristics of the correction staffs data.The main work includes the following points:(1)A clustering algorithm for correcting state information based on feature subspace is proposed.This paper localizes the high-dimensional data search problem in related dimensions to solve the problem of automatic labeling of correction strategy tags.Under the guidance of sociological theory,design data-driven statistical evaluation standards to evaluate the effectiveness of these subspaces.By using the Boost ensemble tree algorithm to search for the importance of features,the clustering patterns of feature subspaces are further discovered,and the case knowledge is decomposed into several sub-cases.Knowledge,so as to improve the accuracy of corrective strategy recommendations.(2)Propose a corrective strategy recommendation algorithm based on deep reinforcement learning.This paper conducts in-depth mining of correction personnel information,constructs correction personnel’s feature set and uses domain knowledge to establish correction strategy evaluation standards,uses improved reinforcement learning algorithm to learn correction personnel characteristics and correction strategy characteristics,and adopts a weighted method.Decision-making fusion realizes the recommendation of correction schemes for correction personnel.(3)Implementation of the correction strategy recommendation system.Through investigation and analysis of the requirements of the system,a corrective strategy recommendation system was built using MVC architecture,front-end and back-end technologies.A layered design is adopted for the system architecture,including the data layer,business layer and user layer.Mysql database is used for the underlying database of data storage,and the front-end webpage is connected with the correction strategy recommendation algorithm through the interface,and finally the system runs stably.This article relies on the national key research and development plan "Research on parole based on social status monitoring big data,and correctional intelligent decision-making technology for temporary provision of out-of-custodial executives",combined with big data correction personnel data collection methods,in the label automatic marking and correction strategy intelligent decision-making technology A breakthrough has been formed to make the formulation of correction plans more intelligent and accurate,promote the solution of problems in the community correction field,optimize the social system structure,and bring good correction benefits.According to the data of the correction personnel in the project,this paper designs experiments to realize the algorithm of this paper,and compares different algorithms.The experimental results show the effectiveness of the algorithm of this paper.