摘要
在推荐系统中用户评分矩阵通常含有大量的遗漏值。这严重影响了协同推荐算法的推荐精度。常用的解决方法是使用缺省值或预测值代替这些遗漏值。通过实验比较了使用不同的替代值的效果,并提出了一种结合矩阵划分和评分预测值的方法。实验结果显示,通过这种方法获得的替代值可以使推荐系统达到更好的推荐质量,尤其是在评分矩阵非常稀疏的情况下。
User＇s rate matrix generally includes many missing values. It seriously debases the precision of collaborative-base recommendation algorithm. The common approach is that using a default value or prediction value to replace the missing value. In this paper we compare the effect of these approaches via experiment and propose an improved approach which combines rate prediction value with matrix partition. Experimental results show that the substitutive value derived from this method can make system get better recommendation quality, especially in the case of sparse rate matrix.
出处
《计算机应用与软件》
CSCD
北大核心
2008年第6期193-195,共3页
Computer Applications and Software
关键词
推荐系统
协同过滤
矩阵划分
评分预测
Recommendation system Collaborative filtering Matrix partition Rate prediction
