摘要
传统协同推荐将用户相似性作为选择推荐者的基准,过多地依赖用户相似性.借鉴Hovland说服模型,提出了基于经验的协同推荐算法.指出推荐效果受多个因素影响,用户经验是选择推荐者时需要考虑的重要因素.给出了从行为日志中测量用户经验的方法,并给出了将用户经验与相似性相结合,整合到标准的协同推荐框架中的两种方法.在真实日志数据上进行了测试.实验表明,与传统方法相比,该方法能够推荐用户感兴趣却意想不到的网页,提高推荐的质量.
Suggesting the most recommendable and trustworthy information that the users can't discover on their own is the just way to improve the quality of recommendation. Traditional collaborative recommendation systems relied heavily on similarity of users. This paper argues that users' similarity alone is not enough and additional factors should also be considered in guiding recommendation. We believe that users' expertise must be an important consideration because people usually trust the suggestions coming from persons with high expertise. We propose a simple method to compute user's expertise from logs and present two ways to incorporate user's expertise into the standard collaborative recommendation frameworks with similarity. We evaluate the approach on a real-world data-set. Experimental results indicate that this method can recommend interesting but unexpected pages to target users and improve the serendipity ratio greatly compared to the existing methods.
出处
《小型微型计算机系统》
CSCD
北大核心
2008年第3期498-502,共5页
Journal of Chinese Computer Systems
基金
北京理工大学基础研究基金项目(0301F18)资助
关键词
协同推荐
个性化
WUM
用户经验
collaborative recommendation
personalization
wUM
User's expertise