论文部分内容阅读
为研究数字电视节目推荐系统不同统计算法的性能,提出利用Rankboost排序算法、Bayes统计算法和简单统计算法三种基于统计模型的算法实现数字电视用户特征的提取与节目推荐。应用实际数字电视运营平台20名用户的测试数据表明,Rankboost算法、Bayes统计算法、简单统计算法排序的AUC(Area Under Curve)值分别为0.732、0.6222和0.6058。分析及测试表明,Rankboost算法因考虑了多重特征在排序中的不同作用,因此在数字电视节目推荐中具有较高的推荐性能。
In order to study the performance of different statistical algorithms in DTV recommendation system, this paper proposes three statistical models based on Rankboost algorithm, Bayesian statistical algorithm and simple statistical algorithm to extract features and recommend programs of DTV users. The test data of 20 users using actual digital TV operation platform shows that the Rank Under, Ranks, Bayes, and simple statistical algorithms have AUC of 0.732, 0.6222 and 0.6058, respectively. Analysis and testing show that the Rankboost algorithm has higher recommendation performance in the recommendation of digital television programs because of considering the different roles of multiple features in ranking.