基于最近邻居聚类的协同过滤推荐算法

Collaborative filtering recommendation algorithm based on nearest neighbor clustering

随着推荐系统用户数量和服务项目增多,可扩展性问题成为推荐算法应用的瓶颈.目前,大部分推荐算法以及基于这些算法的改进主要集中在推荐质量上,随着系统规模扩大,暴露出实时推荐效率降低和运行耗时的缺点.针对这些问题,提出了一种基于最近邻聚类的协同过滤推荐算法.首先,该算法采用二分k-means算法把评分相似的用户划分到相同的类中,以此建立用户聚类模型.然后,从聚类模型中挑选出目标用户的最近邻居类作为检索空间.最后,从检索空间中搜索目标用户的最近邻居,由最近邻居的信息产生最终的推荐列表.实验结果表明,该算法在保持较高的推荐质量的同时可以显著提高推荐系统的效率,比传统的协同过滤算法可扩展性强.

With the increasing number of users and items in recommender systems,designing a scalable algorithm becomes a big challenge for recommendation systems. However, many recommendation algorithms and the improved algorithms proposed thus far have focused on improving recommendation quality,resulting in shortcomings such as lower recommendation efficiency and running time consumption as the system increases in scale. To address the problem of scalability,a collaborative filtering recommendation algorithm based on nearest neighbor clustering was proposed.Firstly,the k-means algorithm was utilized to place similar scores into the same cluster,which was used to build the user clustering model.Then,it picked out the active users' nearest neighbor clusters from the clustering model and treats them as a retrieval space.Finally,the nearest neighbors of an active user are found according to the retrieval space,and the recommendation to the active user was given.Experimental results show that the algorithm proposed in this paper not only significantly improves the response speed of the recommendation system online but also maintains a high accuracy.