Research On Treatment Effect Assessment Of Drug Abusers Based On Federated Graph Learning

Treatment effect assessment refers to the evaluation and analysis of a drug abuser’s abstinence state and the possibility of relapse.Treatment effect of drug abusers is an important standard to assess the efficacy of compulsory isolation detoxification work,and improving treatment effect is the central task of compulsory isolation detoxification work.It is not only related to the vital interests of drug abusers,but also has a very important significance to adjust and improve the measures and means of drug abstinence and further improve the rate of abstinence.At present,most of the work on treatment effect assessment of drug abusers is based on paper scale or traditional machine learning models,which have low efficiency,poor accuracy and insufficient data utilization.At the same time,due to the special nature of drug rehabilitation administration,the management attaches particular importance to the protection of data security and privacy.However,in the actual project implementation,the data interaction security of traditional machine learning models is poor in the training process,and the bandwidth requirement is high.To solve the above problems,this thesis carries out a research based on the application of graph neural network algorithm in tabular data and the application of federated graph learning model in the field of treatment effect assessment.On this basis,the classification algorithm based on node degree influence(GraphSAGE-D)and treatment effect assessment model of drug abusers based on federated graph learning are proposed.The main work of this thesis focuses on the following three aspects:(1)Aiming at the problem that traditional graph neural network algorithm cannot make effective use of node network structure information,this thesis improves on the basis of the GraphSAGE,introduces node’s own feature information and graph structure related information,designs a new aggregation function according to node degree and node influence in the network.A classification algorithm based on the influence of node degree(GraphSAGE-D)is proposed,which solves the problem that the original algorithms only use the attribute information or feature information of the node itself when classifying network nodes,and cannot effectively use the structural information of the node in the network.Comparative experiments and parameter analysis experiments are conducted on public datasets,which prove that the proposed algorithm is more effective,with both macro-F1 and micro-F1 metrics gaining more than 1%improvement over the baseline algorithms on all test datasets.(2)Aiming at the problem that the traditional machine learning models cannot effectively extract the correlation relationship between subjects in the data and the poor security of data interaction and high bandwidth requirement in the training process,this thesis combines with the proposed GraphSAGE-D algorithm,uses Ring All-Reduce algorithm to construct a decentralized federated learning framework,and introduces homomorphic encryption based on Paillier and STC.Treatment effect assessment model of drug abusers based on federated graph learning is proposed.Comparison experiments,ablation experiments and parameter analysis experiments are carried out on the dataset of treatment effect,which prove that compared with the baseline models,the proposed model improves the accuracy of evaluation,as well as the safety and communication performance of data interaction in the training process.(3)Aiming at the problem of low data quality in the original management system,a treatment effect assessment system is constructed by using the treatment effect assessment model proposed in this thesis.The system is mainly composed of four parts:database,model and functional interface,front-end display framework and practical application support.On the one hand,the problem of low data quality in the original management system can be solved by unified data processing.On the other hand,the evaluation function of the system can promote the administration of drug rehabilitation administration to be more scientific and efficient,and at the same time facilitate the integration of data from scattered subordinate agencies in actual projects.