摘要
传统的协同过滤推荐算法以用户对所有物品的评分向量作为计算用户相似度的依据,没有考虑到物品属性对用户兴趣的反映。为此,提出一种新的改进的相似度计算方法,引入了"用户兴趣分布矩阵"的定义,设计了启发式的评分预测方式,即根据兴趣相似度选出TOP-K用户之后,以用户标记的物品数量作为该用户的权重来预测评分。在Movielens数据集上的测试结果表明,改进后的算法相比传统的算法在平均绝对误差(MAE)上降低了7.3%。
The original collaborative filtering recommendation algorithm uses the user’s score vector for all itemsuser＇s similarity,which does not take into account the reflection of the object,s interest to the user’ interest. In this paper,an improved simi-larity calculation methiod is proposed. The newalgorithm introduccs the definition of “ user interest distribution matrix,,,and designs the heuris-tic scoring methiod. The T0P-K users are selected according to the similarity degree of interest,then the number of marked items is used as the weight of the user to predict the score. The results of the test on the Movielens dataset showthat the improved algorithm is 7. 3% lower than the traditional algorithm in the Mean Absolute Error （MAE）.
出处
《微型机与应用》
2017年第15期25-28,共4页
Microcomputer & Its Applications
关键词
协同过滤
兴趣分布
物品属性
用户权重
collaborative filtering
interest distribution
item properties
user weight
