基于Spark的UCSLIM推荐算法研究及实现

Design and Implementation of Parallel UCSLIM Algorithm Based on Spark

稀疏线性(SLIM)推荐算法侧重于通过挖掘物品与物品之间的关系进而产生推荐结果.为了提高推荐质量,借鉴了SLIM算法和协同过滤算法的思想,将用户划分为用户集合,进一步挖掘用户与用户集合之间的隐含关系,并综合考虑用户与用户相关性、用户与用户集合相关性这两个因素,提出了融合用户集合关系的稀疏线性(UCSLIM)推荐算法.实验结果表明,UCSLIM算法能够提高推荐结果质量.同时为了提高算法的执行效率,分别在Spark和Hadoop云计算平台上实现了UCSLIM并行推荐算法,并通过实验表明,UCSLIM的Spark版本具有更高的计算效率.

A top-N recommendation method—user class sparse linear methods（ UCSLIM） based on sparse linear method（ SLIM） was proposed. In order to improve the quality of recommendation,we learn from the idea of SLIM algorithm and collaborative filtering algorithm. The users are divided into different sets. So the correlation was analyzed between the user and the set of users and the correlation between user and user. Based on these two factors,UCSLIM was proposed. Experiments show that,compared with SLIM,UCSLIM can improve the quality of results. Furthermore,in order to improve the computational efficiency,the UCSLIM in Hadoop and Spark was implemented. Experiments show that the implementation by Spark has higher efficiency than that of Hadoop.