基于多源数据融合的协同推荐方法

Collaborating Filtering Method Based on Multiple Data Sources

大数据环境下,实现多源数据的高效利用是推荐服务的一个重要问题。为此,基于协同推荐方法,选取相似用户的消费行为完成推荐排序,在用户聚类过程引入预聚类单元和综合聚类单元。预聚类单元对用户一产品评分矩阵、电信业务身份、用户网络关注内容等数据,分别进行相似度计算。综合聚类单元计算数据丰度权重,进行综合相似度计算和用户聚类。同时,引入了可调节的效度权重,根据推荐结果的点击率和转化率,自动优化推荐系统。

The highly efficient utilization of multiple data sources is a key challenge in big data applications. Based on the collaborative filtering recommendation, services pick consumption behaviors of similar clients by clustering to generate the recommendation list. Client clustering contains two units, one is preliminary clustering, and the other is synthetic clustering. Preliminary clustering use client-product score matrixes, telecommunication service identities, client network behaviors and etc. to calculate similarities. Synthetic clustering weights the abundance of data, and then completes the similarity calculation and client clustering. Adjustable weights of data validity were introduced to optimize the system on the basis of click rates and conversion rates of recommendation list.