摘要： 现有的协同过滤算法未考虑用户浏览记录中用户对项目的潜在厌恶信息,忽视新老用户对不同流行度项目的兴趣差异。为此,提出一种改进的协同过滤算法。从用户浏览记录中提取用户对项目的潜在厌恶信息,计算项目之间被用户厌恶的相似度,将其与项目之间被用户喜欢的相似度结合,得到项目的综合相似度。在此基础上用偏好因子对用户的兴趣度进行缩放,该因子能够反映新老用户对不同流行度项目的倾向性。实验结果表明,该算法在不明显增加时空复杂度的前提下,可有效提高推荐准确率、召回率和覆盖率。
关键词:
协同过滤,
潜在厌恶信息,
偏好因子,
相似度拓展,
兴趣度缩放
Abstract: The existing Collaborative Filtering(CF) recommendation algorithms are not taken into account the latent information of users’ aversion to items in the users’ browsing history and the difference between old users’ interests and new users’ interests in items of different popularity.To solve the problem,this paper proposes a collaborative filtering algorithm based on similarity extension and interest degree scaling.It extracts the latent information of users’ aversion to items from the users’ browsing history to calculate the items’ similarity of user’ aversion,and combines it with items’ similarity of users’ favorite to get the integrated similarity of items.On the basis of that,the algorithm scales the users’ interest degree to the items by preference factor which can reflect the difference between new users’ interests and old users’ interests in both popular items and unpopular items.Experimental results show that this algorithm improves the recommendation precision,recall rate and coverage,without obviously increasing in time-space complexity.
Key words:
Collaborative Filtering(CF),
latent aversion information,
preference factor,
similarity extension,
interest degree scaling
中图分类号: