摘要
协同过滤是推荐系统中应用最为广泛的方法.基于用户的协同过滤算法在计算用户相似性时,对不同的项目给予相同的权重,然而在现实中不同项目对刻画用户的兴趣所起作用不同,从而基于用户的协同过滤会造成对流行的项目打分高的问题,而不能真正反映用户的兴趣.本文提出项目的区分用户偏好值概念,从而更好的刻画了用户的兴趣,在此基础上,改进了计算用户相似度的方法,使推荐算法具有较高准确度.算法在标准数据集MovieLens上进行了测试,实验表明了算法的有效性.
Collaborative filtering is the most widely used method in recommendation systems.In used-based collaborative filtering algorithms,people usually assign the same weight to each item when computing user similarities.However,in reality,different items provide different amount of information for identifying user interests.Correspondingly,the previous way of equally weighting items is unfair and will lead to inaccurate results,e.g.,the used-based algorithms often result in the problem of predicting high scores to the popular items,and thus cannot reflect the given user＇s real preference.To that end,in this paper,we propose the concept of items＇ discriminabilities on user interest for better distinguishing and expressing user interests.Further,based on this concept,we design an enhanced algorithm for calculating user similarities and making recommendations.Since our treatment can reflect user preferences more preciously,the recommendation algorithm gets more accurate results.Extensive experiments on the benchmark MovieLens dataset show the effectiveness of proposed algorithm.
出处
《小型微型计算机系统》
CSCD
北大核心
2012年第7期1533-1536,共4页
Journal of Chinese Computer Systems
关键词
协同过滤
推荐算法
个性化
项目区分用户兴趣度值
collaborative filtering
recommendation algorithm
personalized
items＇ discriminabilities on user interests
