摘要
现有的基于用户的协同过滤推荐算法使用用户—项目评分矩阵计算用户的评分相似性作为用户的相似度,存在矩阵稀疏的问题,而且不能对用户的兴趣进行动态衡量。由此提出一种改进的基于用户的协同过滤推荐算法,通过历史数据计算用户对各类项目的购买数量比例矩阵,衡量用户对各类项目的兴趣;根据用户购买项目的时间的先后衡量用户兴趣的动态变化。融合以上两点得出用户兴趣相似性作为用户相似性的权重,改进的用户相似性计算方法避免了用户—项目评分矩阵的稀疏性和不能动态衡量用户兴趣变化的问题。采用Movie Lens数据集进行实验,结果表明该算法提高了推荐结果的准确性并且具有稳定性。
The existing recommendation algorithm of collaborative filtering based on users, which use user-item rating matrix to calculate users rating similarity, has problem of sparse matrix and cannot dynamically measure user interest. On the basis of this, this paper proposes an improved recommendation algorithm of collaborative filtering based on users, which calculate users＇ propor- tion matrix of purchase quantity for all categories of items through historical data and measure user interest for all kinds of items. The improved recommendation algorithm measures dynamic changes of user interest according to the purchase time of users. On the basis of this, the paper gets the user interest similarity as the weight of user similarity. The improved method of user similarity calculation avoids the sparse of user-item rating matrix as well as the problem that cannot dynamically measure user interest. The paper takes the data sets of MnvieLens for experiment, which the result shows that the algorithm can improve the accuracy and stability of the rec- ommendation result.
出处
《情报理论与实践》
CSSCI
北大核心
2015年第9期100-103,133,共5页
Information Studies:Theory & Application
关键词
协同过滤
用户兴趣
动态兴趣
collaborative filtering
user interest
dynamic interest
