Research On Abnormal Risk Assessment Method Of Specific Population Based On Knowledge Augmentation Graph Neural Network

The prison bears the important responsibility of supervising,educating,and rehabilitating criminals.It is an important institution for safeguarding public safety.In recent years,with the launch of the campaign against organized crime and corruption,as well as the continuous improvement of the social governance system,the proportion of criminals involved in organized crime and corruption in prisons has increased significantly year by year.This has led to an increase in the difficulty of supervision,and incidents of violence,suicide,and escape have become more frequent.Therefore,it is crucial for correctional facilities to achieve objective and accurate risk assessment of inmates,and to scientifically organize risk management.Since the national launch of the inmate risk assessment work,various prisons throughout the country have developed various risk assessment tools.However,there are still some shortcomings,specifically:1)dependence on subjective human experience during risk assessment,resulting in low reliability and validity;2)sparse and heterogeneous data,making traditional statistical analysis-based evaluation methods unable to delve into potential psychological activity;3)a lack of understanding of the psychological and behavioral evolution mechanisms of specific populations,resulting in low accuracy of prediction,lag,and poor targeting.To address these issues,this paper proposes the use of specific population heterogeneous data in prison,as well as technology such as knowledge graph and recommendation system,to study feature fusion representation and modeling methods under psychological/behavioral patterns.Additionally,using models such as graph neural networks,this paper research specific population psychological dynamic laws and spatiotemporal behavior patterns to establish a risk assessment model that can predict unknown risks in advance.The following aspects were mainly researched:(1)To address the problem of the lack of objective correlation between specific population heterogeneous data in prison and the inability to extend the knowledge graph,this paper proposes a specific population knowledge graph construction method based on entity relevance.This method uses weighted entity semantic similarity and relevance feature importance to make decision connections between entities,thereby constructing a specific population knowledge graph that provides rich data foundations and explainable analysis for abnormal risk assessment models.(2)To address the issue of previous graph learning recommendation systems only focusing on a small number of important relationships and not fully utilizing associated entity features,this paper proposes a multi-graph encoder module based on knowledge augmentation graph convolutional networks.This module uses entity relevance to perform adaptive subgraph segmentation,forming multiple relationship subgraphs,and uses graph convolutional networks to parallel process heterogeneous project dependencies on different relationship subgraphs to improve model efficiency and achieve full mining of specific population heterogeneous data.(3)Based on the multi-graph encoder module,this paper further proposes a specific population abnormal risk assessment module based on self-attention mechanism,which can accurately depict the overall characteristics of incarcerated user profiles and psychological behavioral states.This module uses self-attention mechanism-based cross-subgraph embedding learning layers to learn the embedding relevance of multiple relationship subgraphs,enhancing entity representation in the global knowledge context,and finally using a multi-layer perceptron to achieve specific population abnormal risk assessment.This paper uses specific population datasets in prison to conduct ablation experiments and comparative experiments analysis of the proposed knowledge augmentation graph neural network-based specific population abnormal risk assessment algorithm.The experimental results demonstrate the effectiveness and superiority of the proposed algorithm.Through research and demand analysis,the relevant technology was applied to complete the design and implementation of a specific population abnormal risk assessment system.