摘要
个性化推荐系统是应用系统中广泛应用的技术之一,用户兴趣偏好模型的建立与更新是个性化推荐系统的关键环节,针对移动设备位置随时变化的特点,以移动端的应用系统为研究对象,提出了一种随用户位置变化而动态更新的用户兴趣偏好模型,并对实现过程中的几个关键问题,包括用户兴趣偏好模型表示方法、用户兴趣关键字提取、用户兴趣偏好模型的建立与更新算法进行了详细描述,最后利用用户兴趣偏好模型根据协同过滤算法进行个性化推荐,并根据用户对推荐结果的评价进一步修正用户兴趣偏好模型。用户兴趣偏好模型采用基于兴趣关键字的向量空间模型表示,用户兴趣关键字由根据TF-IDF算法获得的用户隐式兴趣和用户参与的显式兴趣相结合获得,用户位置信息变化时,系统获取当前位置附近的服务,对已存在于用户兴趣关键树中的服务权值进行增强,而对不存在其中的进行遗忘以调整用户兴趣树从而更新用户兴趣偏好模型。验证表明,该方法推荐的服务更符合用户所处的位置上下文环境,并且具有高度的可达性。
Personalized recommendation system is one of the technology that is widely used in applications. Establishment of user interest model is the key link of personalized recommendation system for the characteristics of mobile device that changes at any time. A kind of dynamically updated user interest model is represented based on the study of mobile application, and some key questions including representation method of user model, user interest keyword extraction, the algorithm for creating and updating user model are discussed. Finally, collaborative filtering algorithm for personalized recommendation based on the latest model of user interest, and then further amendment user interest model based on user evaluation of the recommendation results. User preference model of interest is expressed by using the vector space model based on the interest keywords. Keywords get attention by combining the implicit interest that get from TF-IDF algorithm and the explicit interest by user involvement. When mobile device detects variation o{ the location, it collects the service that is related to the loca- tion. Then user interest model is updated by strengthening the weights of the service that exists in the user interest tree, re- ducing the weights of the service that does not exist in the user interest tree. Verification shows that the method is more con- sistent with user location context, and with a high degree of accessibility.
出处
《计算机与数字工程》
2014年第9期1571-1576,1640,共7页
Computer & Digital Engineering
基金
北京市属高等学校创新团队建设与教师职业发展计划项目(编号:IDHT20130502)资助
关键词
移动端
个性化服务推荐
用户动态兴趣模型
隐式提取
mobile terminal, personalized service recommendation, dynamic user interest model, implicit extraction