基于用户聚类的播客节目推荐  

USER CLUSTER BASED PROGRAM RECOMMENDATION SYSTEM OF PODCAST

作　　者：陈超[1] 齐开悦[1] 陈剑波[1] 

机构地区：[1]上海交通大学信息安全工程学院,上海200240

出　　处：《计算机应用与软件》2009年第3期7-10,共4页Computer Applications and Software

基　　金：国家973重点基础研究发展计划(2005CB321804)

摘　　要：许多播客推荐机制一般根据整体点击次数来向用户推荐节目,但是一些点击次数很高的节目未必就是某类用户所喜欢的,因而推荐的节目对用户的喜好针对性不是很强。为了提高推荐节目对用户喜好特点的针对性,提出基于用户聚类的节目推荐。对用户在播客平台上的采集数据进行聚类分析后,把用户归为某一类型,并把该类型的所有节目根据点击次数多少存放入相应的推荐表。在用户下次登录时,根据其所属用户类型从推荐表中选出其最可能观看的尚未浏览的节目。试验结果证明,该播客推荐系统能很好地根据用户的喜好特点来进行针对性节目推荐。The basic means of recommending mechanism of podcast is based on the total clicks. However, the programs recommended by such a mechanism would not be those that preferred by certain cluster of users,which means that the recommended programs are not so close to users＇ preference. In order to improve the pertinence of recommendation to fit the preference feature of users,this paper puts up with a new mechanism based on user clustering, which applies cluster analysis on the user data collected on the platform of Podcast and classifies the users into different clusters, and all the programs in a certain cluster will be stored in correlative recommendation table according to their click numbers, thus when users log on later, they may choose those programs that have not been browsed but most possibly to watch from the recommendation tables in relevant users＇ cluster they belong to. The results showed that the new podcast recommendation system is well capable of recommending the pertinency programs in line with users＇ preference.

关 键 词：播客 用户聚类 节目推荐 

分 类 号：TP393[自动化与计算机技术—计算机应用技术] TN948.14[自动化与计算机技术—计算机科学与技术]