基于Hadoop的改进协同过滤算法研究

RESEARCH OF LMPROVED COLLABORATIVE FILTERING RECOMMENDATION ALGORITHM BASED ON HADOOP

协同过滤算法是个性化推荐系统中一种常用算法,但实际应用时还存在着数据稀疏性,用户兴趣变化和扩展性等问题。针对稀疏性问题,采用预测评分策略填充原始矩阵,减小稀疏性对算法精度的影响。针对用户兴趣动态变化问题,把时间因素的影响考虑进去,改变不同时刻评分的权重,解决对用户兴趣更新不及时所导致的推荐结果不够全面、准确的问题。针对算法在面临大数据处理时的扩展性问题,将改进后的算法移植到云平台,利用Hadoop在分布式计算、存储等方面的优势解决。实验结果证明,改进后的算法对上述算法得到较好的解决。

Collaborative filtering is one of the most often used algorithms of personalized recommendation system,but there are issues of data sparsity,user interests change,and extensibilityin actual applications. For data sparsity,the improved algorithm fills the original matrix by adopting the strategy of predicted ratings,in order to reduce the effect on its the accuracy. In addition,for the dynamical user interest change,the improved algorithm takes time factor into consideration,and changes the weight of different time in order to solve the problems that are not comprehensive and accurate for recommendation results,which is caused by user interests that are not updated in time. Finally by taking advantage of its distributed computing and storing,we put the improved algorithm on cloud platform to solve extensibilityproblem. The experiment results show that the improved algorithm can solve these problems better.