Research And Implementation Of Distributed Union Scorecard System For Data Privacy

Scoring card is a widely used scoring and rating model,and due to its excellent interpretability,it is still widely used in financial and medical scenarios even today when deep learning is prevalent.However,with the continuous development of technology,massive data generation and the release of laws related to data security and data privacy protection in various countries,the establishment of cross-organizational union scorecards is called an inevitable trend.In the face of such a demand,the following problems exist in the current research practice of scorecard systems:1)the lack of data privacy protection in the process of score calculation,the contradiction between the large-scale multi-dimensional data demand of scorecard models and the increasingly difficult circulation of privacy data;2)the known privacy calculation techniques have a large overhead and many limitations;3)scorecard modeling is dominated by private data,and there is a lack of a one-stop platform for cross-organizational scorecard A one-stop platform for scorecard construction and evaluation.In the face of the above problems and needs,the main research of this paper is as follows:(1)To address the issue of securing data privacy in the scoring computation process,this paper proposes a reference architecture for privacy scoring computation.The architecture introduces edge nodes to improve the access to data,and each participant is only responsible for storing and processing the data in its possession,and ensures data security and privacy protection through encryption and secure communication protocols.In addition,a security verification scheme for scorecard computation models based on surrogate method and statistical deviation is proposed and used to improve data privacy guarantees when dealing with malicious models.It makes it possible to complete the score calculation with data privacy protection.(2)To address the overhead problem of privacy computing techniques,this paper proposes an efficient privacy computing scheduling scheme.Through recursion-based task splitting,automatic scheduling and corresponding auxiliary strategies,we achieve maximum parallel distributed computation and trade space for time.Moreover,it relies on an independent privacy security module to achieve reliable decryption and noise reduction of ciphertexts and solve the hierarchical multiplication limitation of homomorphic encryption,which can reduce the ciphertext volume expansion to a certain extent.(3)For cross-organizational implementation of scorecards,this paper researches and implements a distributed federated scorecard platform for data privacy,providing a cross-organizational scorecard construction and evaluation visualization platform.It includes cross-organizational data access and authorization,union modeling and data evaluation,and other functions.It solves the problem of accessing and evaluating the use of cross-organizational data,and provides a good experience and support for the analysis and research work of various businesses of individuals or enterprises.Through the research of this thesis,an effective reference is provided for the cross-organizational implementation of existing scorecard systems.