基于用户兴趣的动态近邻协同过滤算法

Collaborative Filtering Algorithm for Dynamic Neighborhood Based on User Interest

为了帮助人们从大量互联网资源中找到感兴趣的信息,推荐系统由此而生.其中,应用最广泛,也是最早出现的推荐算法包括协同过滤算法,但是该算法还存在着许多不足之处.该算法主要考虑用户的评分数据,未能结合项目进行考虑,同时在选取当前用户的最近邻用户时,通常统一规定了近邻用户数目,没有结合每个用户的实际数据,导致推荐的效果无法取得最优.因此,本文在充分考虑用户评分的情况下,还结合项目信息加入了用户的兴趣偏好,提出了一种基于用户兴趣的动态近邻协同过滤算法.综合用户的标签数据和评分数据来计算相似度,可以很好的缓解仅依靠评分数据带来的稀疏性问题.同时在得到用户之间的相似度之后,设定2个阀值,分布选取最近邻用户.只有当用户间相似度超过阈值,该用户才会被选择为最近邻的用户,动态的找到每一个用户的严格最近邻用户.通过实验,与常用的协同过滤算法相比,本文提出的算法推荐的误差更小,并且为以后的研究工作奠定了基础.

The recommended system help users to find what they want in a large amount of information.Collaborative filtering is one of the most widely used and oldest recommended technology. However,the algorithm mainly considers the score between users,fails to fully consider the labels of the items,and the user nearest neighbor number is unified,so the recommended result is not very good. A dynamic neighbor of collaborative filtering algorithm,based on items labels,was proposed,and the similarity between users was calculated by rates and labels with their respective weight,the similarity algorithm effectively reduced the sparsity of the similarity matrix of user rates. At the same time,after calculating the similarity of the user,by setting the similarity threshold,the nearest neighbor users of each user was dynamically found. At this time,only the similarity was greater than the similarity threshold,which was the nearest neighbor. Compared with the user-based collaborative filtering algorithm,it was verified that the algorithm got better recommendation results and layed the foundation for further research work.