数字电视节目推荐系统中的统计算法

Statistical algorithms for digital TV program recommendation systems

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。

The performances of various statistical approaches for digital TV program recommendation systems were analyzed using the Rankboost algorithm, the Bayesian statistical algorithm, and a simple statistical algorithm for user feature extraction and program recommendation. Comparisons with the recommendations of 20 viewers of actual cable digital TV show that the AUC of the Rankboost algorithm was 0. 732, that of the Bayesian statistical algorithm was 0. 622 2, and that of the simple statistical algorithm was 0. 605 8. Theory analyses and tests show that the Rankboost algorithm has the best performance for digital TV program recommendations because it considers the more influences of multiple features in the ranking compared with the other two algorithms.