Research And Implementation Of Community Correction Strategy Recommendation Technology Based On Multi-model Fusion

With the popularization of community correction policies across the country,information on correction personnel are increasing.At the same time,as the community correction institutions are gradually improved,more and more factors of correction personnel are considered during the correction process.The analysis of wide-scale and fine-grained information has brought some pressure to the community correction work,which is reflected in how to recommend reasonable correction strategies for correction staff.This paper focuses on the personalized recommendation of community correction strategies,analyzes the true information of community correction personnel,and uses multiple model fusion methods to conduct research.The main research contents of this paper are as follows:(1)Combined with the knowledge of community correction field,this paper analyzes the characteristics of community correction workers' portraits.According to the data set,an improved information compression feature encoding and semantic encoding method based on the Bert language model are proposed.The importance of features is evaluated by random forest,and the current popular boost integrated tree algorithm is used to predict.Finally,the optimal model combination is obtained through comparative experiments.(2)This paper deeply excavates the characteristics of community correction personnel,introduces feature of community correction strategy information,and proposes an improved neural network structure based on attention mechanism,which combines the static feature of correction personnel's own attributes and the dynamic behavioral psychological feature with the feature of community correction strategy,and finally outputs recommended results.The results show that the proposed algorithm has better recommendation performance.(3)The advantages and disadvantages of two algorithms are analyzed.The two algorithms are combined by multi model fusion,which reduces the cold start problem,improves the recommendation accuracy and the generalization ability of the model.the correction strategy recommendation module is designed to realize the front-end visual display of correction strategy recommendation.