基于时段-时长耦合LDA的用户收视行为挖掘

网络协议电视(IPTV)的用户收视兴趣不仅体现在用户观看的节目列表,还体现在节目的观看时间点和时长上。考虑到现有方法对时间点和时长的忽略,提出一种时段-时长耦合的LDA模型。通过刻画用户兴趣主题和收视时段的隐变量生成收视记录中的观看节目、观看时间点和时长,并用Gibbs采样对上述隐变量进行推断。在天津电视台用户行为数据上进行验证,结果表明,该模型可以得到节目相关性更高的兴趣主题,更加精确地挖掘到用户在不同时段的收视兴趣分布。将该模型用于IPTV节目推荐,相较于传统的cLDA,推荐效果有显著提升。

TIMEK-DURATION COUPLED LDA FOR USER VIEWING BEHAVIOR MINING

The interests of IPTV(Internet Protocol Television)user is not only reflected in the program list viewed by users,but also in the time and duration of watching programs.Considering the ignorance of time and duration in existing methods,this paper proposes a time-duration coupled latent dirichlet allocation(TDC-LDA)model.The probability model generated a viewing program,a viewing time point and a duration in the viewing record by characterizing the hidden variables of the user interest topic and the viewing period,and inferring the hidden variable by Gibbs sampling.It was verified on the user behavior data of Tianjin TV station.The experimental results show that the model can obtain the interest topic with higher program relevance,and more accurately mine the user s viewing interest distribution in different time periods.Compared with the traditional cLDA,the proposed model is more effective on IPTV program recommendation.