【摘要】 为提高推荐精确度,提出了一种基于用户偏好的矩阵分解推荐算法(USPMF)。综合考虑通过对用户之间的相似性、用户与项目之间的信息的分析,同时考虑数据量大引起的时间和空间复杂度高的问题,引入了矩阵分解方式。USPMF算法以优化损失函数为目标,在达到全局最优的同时,提高预测的准确度。将USPMF算法与正则化矩阵分解算法、基于用户的协同过滤推荐算法进行了比较,在真实的数据集上的实验结果表明,USPMF算法在预测准确性上有显著提高,平均绝对误差(MAE)分别降低了13.70%、1.17%,均方根误差(RMSE)分别降低了15.07%、1.03%。更多还原

【Abstract】 To improve the accuracy of recommendation, the paper proposed USPMF, a matrix factorization recommendation algorithm based one users ’ preference. Based on the analysis of the similarity between users and the information between users and items while considering the problem of high complexity of time and space caused by large amount of data,matrix factorization was introduced. The USPMF algorithm optimized the loss function as the goal to achieve global optimization while improving the prediction accuracy. USPMF algorithm was compared with regularized singular value decomposition recommender algorithm and user-based collaborative filtering algorithm. The experiment results on a real dataset show that USPMF algorithm has significant advantages in the accuracy of forecast. The Mean Absolute Error( MAE) is decreased by 13. 7%,1. 03% and Root Mean Squared Error( RMSE) is decreased by 15. 7%,1. 03%.更多还原