【摘要】 随着直播频道的不断增加,观众不得不花费额外的时间和精力来选择适合的节目.通常,推荐系统可有效缓解上述问题,然而直播电视所具有的实时性、多用户、隐性反馈、冷启动等特点给推荐系统的研究带来挑战.针对这些特点,现有的方法大多利用时段划分的方式将用户对节目的偏好转换为对频道的偏好,通过推荐频道来完成节目的推荐.然而,这些方法的时段划分规则依赖经验,不具通用性,可解释性差,并且未考虑用户偏好会随时间的推移而变化的情况,同时,推荐频道的模型往往忽视了对正在播出的节目的关注.为此,本文提出了时间相关的直播电视推荐算法TTRec.首先,采用聚类的方法按时段将每个物理频道划分为若干虚拟频道,构建用户-虚拟频道偏好矩阵.其次,利用遗忘函数调整用户隐性反馈的量化结果,将其作为偏好矩阵的填充值.然后,采用协同过滤算法对偏好矩阵中未知项进行预测.最后,以节目的时间特征为属性,通过计算正在播出的新节目与历史节目的相似度来调整偏好矩阵中对应的数值,按调整后的结果生成推荐列表.在真实的数据集中的对比实验表明,TTRec明显优于对比算法.更多还原

【Abstract】 With the increasing of live TV channels,viewers have to spend extra time and energy to choose suitable TV shows.Generally,a recommender system is able to effectively alleviate this problem,but the real-time-multi-user, implicit feedback,cold start bring challenges to the recommender systems of live TV.Facing these characteristics,most of the methods adopted time-division strategies to shift the user’s preference on programs to the user’s preference on TV channels,and recommend channels to perform the program recommendation.However,these time-division strategies rely on experience,are not universal,have poor interpretability,and do not take into account the fact that user preferences will be changed with the passage of time.Meanwhile,those channel-recommended models ignore the characteristics of on-aired programs.Therefore,we propose a time-dependent recommendation algorithm for live TV,TTRec.Firstly,each physical TV channel is divided into several virtual channels by clustering,and a user-virtual channel preference matrix is constructed.Secondly,memory functions are used to adjust the evaluation of implicit feedbacks,which are used as filling values of preference matrix.Then,we adopt collaborative filtering algorithm to predict the unknowns of preference matrix.Finally,the corresponding values in the preference matrix are adjusted by the similarity between the on-aired program and the historical programs,and a recommendation list is generated according to the adjusted results.Comparison experiments in real data sets show that TTRec is superior to the comparison algorithms.更多还原