智能传播时代下算法推荐的失控与重构  

Dilemma and Reconstruction of Algorithm Recommendation in the Era of Intelligent Communication

作　　者：薛可 李亦飞[1] XUE Ke;LI Yifei(School of Media and Communication,Shanghai Jiao Tong University,Shanghai 200240,China;China Institute of Urban Governance,Shanghai Jiao Tong University,Shanghai 200030,China)

机构地区：[1]上海交通大学媒体与传播学院,上海200240 [2]上海交通大学中国城市治理研究院,上海200030

出　　处：《上海交通大学学报（哲学社会科学版）》2023年第5期22-37,共16页Journal of Shanghai Jiao tong University(Philosophy and Social Sciences)

摘　　要：算法推荐的广泛应用改变了网络信息议程设置的根本逻辑,它通过“内容评分”与“用户聚类”构建信息的分发机制实现精准传播。本研究基于规模性社会调查,发现以流量为导向的算法推荐系统引发了网络信息传播秩序的失控,带来了公众价值偏差、网络集群行为等系列社会问题。以社会问题为导向,本文基于社会责任理论,在弱监督学习的大框架下引入迁移学习方法,通过构建文本-情绪分析模型,将内容正能量属性、网络情绪倾向、评论情感互动共同纳入对算法推荐的优化重构系统,并凭借虚拟仿真进行了验证性分析。研究发现,内容正能量属性的定量判断,可使社会责任对算法技术产生从理论指导到实践优化的效能,成为算法社会责任的履责新范式;内容情感分析的置入可使算法推荐形成对网络信息内容更客观、更完整的评估结论,实现对舆情事件的监测;将评论情感互动作为影响网络集群行为的重要因子纳入算法评估体系,则能够深度把控情绪传播的模仿-反馈机制,最终实现对集群行为的预警,维护网络生态环境的良性发展。The widespread use of algorithm recommendation has changed the fundamental logic of agenda setting.The distribution mechanism of information on the internet is constructed by“content scoring”and“user clustering”to achieve accurate dissemination.Based on a large-scale social survey,this paper finds that the flow-oriented recommender system has triggered the uncontrolled order of network information dissemination and brought about a series of social problems,such as the public value bias and internet collective behavior.Guided by social issues,this paper introduces transfer learning methods under the general framework of weakly supervised learning based on the social responsibility theory.By constructing a text-emotion analysis model,it jointly incorporates the positive content attributes,network emotional tendencies,and comment emotional interactions into an optimal reconstruction system for algorithm recommendation,and conducts a validation analysis with virtual simulation.The findings indicates that the quantitative judgment of positive content attributes can realize the effectiveness of social responsibility on algorithm recommendation from theoretical guidance to practical optimization and become a new paradigm for algorithm social responsibility.The placement of content emotion analysis enables the recommender system to form more objective and complete evaluation conclusions on the content of online information and to realize the implementation of monitoring public opinion events.Incorporating comment emotional interactions as an important factor influencing internet collective behavior into the algorithm evaluation system can deeply control the imitation-feedback mechanism of emotion propagation,ultimately achieving early warning of collective behavior and maintaining the benign development of the network ecological environment.

关 键 词：算法推荐 社会责任 网络情绪 网络集群行为 文本-情绪分析模型 

分 类 号：G206[文化科学—传播学] TP391.3[自动化与计算机技术—计算机应用技术]