The rapid development of mobile Internet brought huge user behavior logs to online reading system. In the face of in-creasingly large terabytes even petabytes user log data, we design a user model and user clustering solution based on hive ware-house. This method can accurately depict the various dimensions and scales of user preferences, building dynamic user require-ment model based on user reading behavior and apply clustering algorihtm to divide user into groups to provide service for per-sonalized Web application, such as recommendation, search, advertisement delivery. Test results show that this method can take advantage of the storage and computing power of hadoop cluster, thus has a good performance and speed of execution.%移动互联网的高速发展为在线阅读系统带来了海量的用户行为日志.针对日益巨大的TB甚至PB级用户行为日志数据,该文设计一种基于Hive数据仓库的用户模型及用户聚类方案.该方法能够准确的基于用户的阅读行为刻画用户的多维度、多尺度偏好特征,构建动态用户需求模型,并基于用户特征进行聚类,划分用户集,为个性化推荐、搜索或者广告投放等Web个性化应用提供服务.实验结果表明,该方法可以发挥集群存储和运算的优势,具有良好的性能和执行速度.
展开▼
