基于大语言模型的短视频平台生态中的用户破圈分析——以快手为例  

Analysis of User Circle-Breaking Behavior in the Short Video Platform Ecology Based on Large Language Models:A Case Study of Kuaishou

作　　者：潘禹辰 杨紫婷 汤昊天 麦柏荣 鲁提普拉·卢合曼 徐璐 Pan Yuchen;Yang Ziting;Tang Haotian;Mai Bairong;Lutipula Luheman;Xu Lu(School of Information Resource Management,Renmin University of China,Beijing 100872)

机构地区：[1]中国人民大学信息资源管理学院,北京100872

出　　处：《图书情报工作》2025年第4期34-52,共19页Library and Information Service

基　　金：国家自然科学基金面上项目“信息‘精炼’下基于时空网络演化分析的多元O2O服务推荐方法研究”(项目编号:72471231);中国人民大学科学研究基金项目(中央高校基本科研业务费专项资金资助)“个性化推荐的‘卡脖子’难题研究—在O2O商业领域的应用”(项目编号:24XNR07)研究成果之一。

摘　　要：[目的/意义]随着短视频行业的迅猛发展和用户规模的不断扩张,短视频平台广泛采用的个性化推荐算法在提升用户体验的同时,也带来用户信息同质化的问题,进而限制用户信息获取的广度和深度。深入探究用户在短视频平台上的破圈行为,即用户如何突破其固有的视频类型偏好圈,以期为用户提供更加多元化、全面化的信息获取体验,并进而促进短视频平台的健康发展。[方法/过程]结合理论建构与实证研究的方法,选取快手短视频平台作为研究案例。提出用户行为分析AIS模型,并基于此模型对用户群体进行细分,进而分析不同用户群体的破圈行为和破圈意愿。同时,利用大语言模型技术,设计一个辅助用户破圈的推荐系统架构,以期通过技术手段促进用户的信息多元化探索。[结果/结论]快手短视频平台的用户可被细分为4类,这些用户群体在视频消费和社交互动方面展现出差异化的特征。总体而言,用户离开原有视频观看偏好圈、主动选择观看新视频的意愿普遍较低。然而,值得注意的是,那些在社交网络上表现活跃,但对具体内容的消费和互动并不频繁的用户,反而更容易发生破圈行为。因此,借助平台和算法的力量来辅助用户破圈,是提升用户体验和促进平台健康发展的重要途径。[Purpose/Significance]With the rapid growth of short videos and the increasing number of users,the personalized recommendation algorithms widely adopted by short video platforms improves the user experience,but also brings the problem of users’information homogenization,which in turn restricts the breadth and depth of users’information acquisition.The purpose of this study is to explore users’circle-breaking behavior on short video platforms,specifically how users break through their existing video preferences,in order to provide users with a more diversified and comprehensive information acquisition experience,and promote the healthy development of short video platforms.[Method/Process]This study combined the method of theoretical construction and empirical research,taking Kuaishou short video platform as a case study.The AIS model of user behavior analysis was proposed,which was used to segment users and analyze their circle-breaking behaviors and willingness.At the same time,based on the large language model technology,a recommendation system architecture was designed to assist users to break the circle,aiming to promote users’information diverse exploration by technical means.[Result/Conclusion]Kuaishou users can be subdivided into four categories.Each of them shows different characteristics in their video consumption and social interaction.Overall,users show a low willingness to leave their original preferred videos and actively choose new ones.However,it is worth noting that users who are active on social networks but do not frequently consume or interact with specific content are more likely to exhibit circle-breaking behavior.Therefore,leveraging platforms and algorithms to assist users in breaking the circle is an important way to enhance user experience and promote the healthy development of platforms.

关 键 词：短视频 用户破圈行为 个性化推荐 大语言模型 

分 类 号：G206[文化科学—传播学] G252