基于Hadoop的多特征协同过滤算法研究

Research on multi-feature collaborative filtering algorithm based on Hadoop

协同过滤是互联网推荐系统的核心技术,针对协同过滤推荐算法中推荐精度和推荐效率以及数据可扩展性问题,采用灰色关联相似度,设计和实现了一种基于Hadoop的多特征协同过滤推荐算法,使用贝叶斯概率对用户特征属性进行分析,根据分析结果形成用户最近邻居集合,通过Hadoop中的MapReduce模型构建预测评分矩阵,最后基于邻居集和用户灰色关联度形成推荐列表。实验结果表明,该算法提高了推荐的有效性和准确度,且能有效支持较大数据集。

Collaborative filtering is the critical technology for Internet recommendation system. In order to deal with the precision,efficiency and the scalability of data set in the recommendation algorithm,this paper proposed a multi-feature collaborative filtering algorithm with grey correlation similarity based on Hadoop architecture. With the analytic result of users’ features based on Bayesian probability model,it formed the effective nearest neighbors’ set. It built the prediction score matrix with the MapReduce model of Hadoop. At last,it formed a recommendation list with the grey correlation similarity model based on the effective nearest neighbors’ set and the prediction score matrix. It improved experimental results show that the efficiency and precision of recommendation. And this algorithm is able to deal with the large-scale data set.