基于改进RFM模型的直播平台用户细分及个性化推荐方法研究  

Research on User Segmentation and Personalized Recommendation Method for Live-streaming Platform Based on Improved RFM Model

作　　者：席运江[1,3] 郭黛翎 廖晓 杜蝶蝶 廖开际 XI Yunjiang;GUO Dailing;LIAO Xiao;DU Diedie;LIAO Kaiji(School of Business Administration,South China University of Technology,Guangzhou 510641,China;School of Internet Finance and Information Engineering,Guangdong University of Finance,Guangzhou 510521,China;School of Management,Guangzhou City University of Technology,Guangzhou 510800,China)

机构地区：[1]华南理工大学工商管理学院,广州510641 [2]广东金融学院互联网金融与信息工程学院,广州510521 [3]广州城市理工学院管理学院,广州510800

出　　处：《竞争情报》2022年第3期36-47,共12页Competitive Intelligence

基　　金：国家自然科学基金项目“虚拟健康社区信息可信度评价模型及智能推荐研究”(编号:72171090)的研究成果之一。

摘　　要：网络直播是一种新兴的共情式传播媒体。针对直播行业特点研究了适合直播平台的用户细分及个性化推荐方法。基于直播平台中用户粘性及长期价值的重要性,将用户平均观看时长作为重要维度融入现有RFM模型,构建RFMT模型;通过RFMT Cluster和RFMT Segmentation方法实现用户细分;通过指标加权法计算用户-主播价值偏好,构建用户评分矩阵,结合协同过滤算法开展推荐实验。基于国内某主流直播平台的后台数据进行研究,结果表明,基于RFMT Segmentation的用户细分方法更适用于实际应用的推荐,细分后的各类群体推荐效果大幅提升。RFMT模型增加了对用户粘性和长期价值的考虑,更适应直播平台的特点,所提出的基于直播平台的用户细分及个性化推荐方法有效、可行。Live-streaming platform is a new kind of empathic media.According to the characteristics of the live-streaming industry,this paper studies the user segmentation and personalized recommendation methods suitable for the live-streaming platform.Based on the importance of user stickiness and long-term value in the live-streaming platform,the average viewing time of users is integrated into the existing RFM model as an important dimension to build the RFMT model.On this foundation,we propose RFMT Cluster and RFMT Segmentation to segment users with different values.The weighted method of index is used to calculate the user-anchor value preference,and the user rating matrix is constructed based on it.Combined with the collaborative filtering algorithm,the recommendation experiment is carried out.This paper is based on the background data of a domestic mainstream live-streaming platform.The verification results show that the user segmentation method based on RFMT Segmentation is more suitable for practical application recommendations,and recommendation effect of various groups after subdivision has been greatly improved.RFMT model increases the consideration of user stickiness and long-term value,which is more suitable for the characteristics of livestreaming platform.The proposed method of user segmentation and personalized recommendation based on live-streaming platform is effective and feasible.

关 键 词：直播 RFM模型 RFMT模型 用户细分 个性化推荐 

分 类 号：TN948.1[电子电信—信号与信息处理] TP391.3[电子电信—信息与通信工程]