基于时间加权标签的协同过滤推荐算法研究  

The Research in Collaborative Filtering Recommendation Algorithm Based on Time-tags

作　　者：宋伟伟[1,2] 杨德刚[1] 郑敏[1] 

机构地区：[1]重庆师范大学计算机与信息科学学院,重庆401331 [2]河南工业与贸易职业学院,郑州450053

出　　处：《重庆师范大学学报（自然科学版）》2016年第5期113-120,共8页Journal of Chongqing Normal University:Natural Science

基　　金：国家自然科学基金(No.10971240);重庆市自然科学基金(No.cstc2012jjA40052;No.cstc2013jcyjA0973;No.cstc2013jcyjA80013);重庆市教委科学技术研究项目(No.KJ120615;No.KJ120630;No.KJ130611;No.KJ1400505);重庆师范大学校级项目(No.13XLZ01;No.201334;No.xyjg13010);校级研究生科研创新项目(No.YKC15009)

摘　　要：传统的协同过滤推荐算法主要以评分数据为数据源来计算和反映用户的兴趣偏好和资源相似度,并决定是否对潜在用户进行推荐物品。而忽视了用户、资源本身的特征,用户在不同时间对资源的认识和感兴趣程度是会变的。基于这个问题,本研究对传统算法进行了改进,提出了基于时间加权标签的信息推荐算法。该算法的主要思想是标签可被用户依个人偏好进行资源标注,标签代表用户对资源的兴趣特征,以用户集、时间、标签集及物品资源等4个量形成的多维关系,可以计算出用户和资源之间的标签特征向量,计算在不同时间段,用户对资源的偏好以及资源相似度,并且依据用户的历史行为来预测用户的偏好,并进行推荐。实验结果显示本算法有效地提高了推荐的准确性,获得了更好的推荐效果。The traditional collaborative-filtering algorithm is based on users-rating data resource to reflect user＇s interests and calculate the resources similarity,the aim is to decide whether to recommend items to potential users or not.But it ignored the characteristics of users and resources,the users＇understanding and interests for the resources may be changed at different times.To solve this problem,this paper puts forward the recommendation algorithm which is based on time-tag information.The main idea is that labels can be tagged by the users according to their preferences for free resources,it reflect the users＇interests and the resources characteristics information.According to the multidimensional relationship of the users,time,tags,and sources,which can generate the tags-feature-vectors of users and resources,then calculate the user＇s interests and the resources＇similarity in different time,the aim is to predict users＇preferences for the further prediction and appropriate recommendation.The experimental results show that it can improve the recommendation accuracy,and obtain better recommendation results.

关 键 词：时间标签 推荐 个性化 准确性 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]