社交平台算法风险内容特征及其贝叶斯关联性

Content Features and Bayesian Correlations of Social Platform Algorithm Risks

[目的/意义]算法风险治理是国家总体安全观的重要组成部分,基于主体感知视角识别算法风险结构及关联,能够为算法风险的防范治理提供参考借鉴。[方法/过程]基于感知风险理论,结合902份深度访谈和微博评论混合数据,扎根构建社交平台用户感知算法风险结构模型,并对其关联性展开贝叶斯复杂网络分析。[结果/结论]感知算法风险涵盖算法自身技术风险和算法外延社会风险两个维度8类风险,其中,算法操纵风险是感知算法风险的核心维度,算法共谋风险和算法黑箱风险、算法致瘾风险的关联关系最紧密;信息质量缺陷和行为操纵是关键节点,算法操纵风险以行为操纵为主;社交平台算法应用中存在“算法悖论”现象,即用户算法认知与算法态度间存在背离。该研究完善了现有算法风险理论框架。

[Purpose/significance]Algorithmic risk governance is an important part of the overall national security concept.Identifying the structure and correlation of algorithmic risk based on the perspective of subject perception can provide reference for the algorithmic risk governance.[Method/process]Based on the theory of perceived risk,combined with the mixed data of 902 in-depth interviews and microblog reviews,the risk structure model of social platform user perception algorithm is built,and the Bayesian complex network analysis is carried out on its relevance.[Result/conclusion]Perceived algorithm risk covers 8 types of risks,including the technical risk of algorithm itself and the social risk of algorithm extension.Algorithm manipulation risk is the core dimension of perceived algorithm risk,and algorithm collusion risk is the most closely related to algorithm black box risk and algorithm addiction risk.Information quality defect and behavior manipulation are the key nodes,and the risk of algorithm manipulation is mainly behavior manipulation.There is an“algorithm paradox”phenomenon in the application of algorithms on social platforms.This research improves the existing theoretical framework of algorithmic risk.