摘要
基于矩阵分解模型、时间因素和排名模式,提出一种局部协同过滤的排名推荐算法,并放松用户对项目的评分矩阵是低秩的这一假设,假设用户对项目的评分矩阵是局部低秩的,即评分矩阵在某个用户项目序偶的近邻空间内是低秩的。修改信息检索中常用的评价指标平均倒数排名(mean reciprocal rank,MRR)函数,使其适合评分数据集合,然后对其进行平滑化操作和简化操作,最后直接优化这一评价指标。提出的算法易于并行化,可以在大型的真实数据集合上运行。试验结果表明该算法能提升推荐的性能。
Based on matrix factorization model,time factor and ranking problem,a collaborative filtering algorithm was proposed. The method relaxed the low-rank assumption of rating matrix and assumed that the rating matrix was locally low-rank,which meaned that the rating matrix was low-rank in the neighborhood of certain user-item combination.Mean reciprocal rank（ MRR）,an evaluation metric widely used in Information retrieval,was modified to fit the rating dataset. The evaluation metric was smoothed and simplied,and then was optimized. The algorithm was easy to parallelize and could operate on real data set. Experiments showed that this algorithm could improve recommendation performance.
出处
《山东大学学报（工学版）》
CAS
北大核心
2016年第5期29-36,共8页
Journal of Shandong University（Engineering Science）
基金
国家自然科学基金资助项目(61370130)
北京市自然科学基金资助项目(4142042)
关键词
推荐系统
协同过滤
矩阵分解
时间因素
平均倒数排名
recommendation system
collaborative filtering
matrix factorization
time factor
mean reciprocal rank
