偏好强度可调的在线直播平台用户交互机制设计及实证研究

A Preference Adjustable Mechanism for Viewer-Streamer Interactions

在线直播平台的大规模交互数据为人类行为的定量分析提供了平台和数据基础,为大规模人群的在线交互机制挖掘提供了难得的契机.文章以斗鱼直播平台2020全年涉及190多万主播的直播大数据为分析基础,基于二部图理论构建了观众-主播交互网络,通过设计用户的交互机制提出了偏好强度可调的观众-主播交互网络演化模型.经实验验证,模型能够准确预测观众-主播交互网络在未来1至8天的演化结果.文章揭示了网络直播平台观众-主播交互过程中观众对直播间已有观众数量的强偏好性,直播间已有的观众数量越多被选择的概率越大,反映了人类对有价值内容的趋向和偏好,印证了真实社会系统中声誉或口碑的累积效应,为网络直播大规模在线人群交互行为特性和内在机理挖掘提供了量化仿真模型和实证分析结果.同时文章为大规模在线人群交互行为特性和内在机理挖掘提供了泛化性较强的可迁移研究框架,对利用大规模在线社交网络的连边和演化规律来描述并预测社交关系的形成和发展过程有重要意义.

The large-scale interaction data of the online live streaming platform provides experimental datasets for the quantitative analysis of human behavior,and offers a new opportunity for the mining of the online interaction mechanism with collective dynamics.Given the lack of empirical research on real-time collective interaction,this paper collects a one-year-long comprehensive dataset of real-time live streaming statistics,involving more than 1.9 million streamers from Douyu(the largest live streaming platform in China),and designs a generalized evolution model for exploring the interaction mechanism between streamers and viewers.First,we construct a viewer-streamer bipartite interaction network representing the dynamics of the entities in the platform,and then propose an evolution model with adjustable preference strength of viewer-streamer interaction.The preference strength can be adjusted with two parameters:The fraction of random choice and the preference coefficient of viewers.Experiments on empirical datasets show that the model can accurately and robustly predict the evolution process when all viewers have linear preference on the number of existing viewers attracted by the streamer when they select a streamer to interact with.This paper reveals the dominating mechanism of preferential attachment for the viewers selecting a streamer and reflects the human tendency and preference for valuable content,confirming the cumulative effect of reputation or word-of-mouth in social systems.Our study provides a quantitative model for exploring the interactive behavior characteristics and internal mechanism of large-scale online crowds in live streaming,and is of great significance for describing and predicting the formation and development of social relationships in more general settings.