基于高阶偏差的因子分解机推荐算法

发布时间：2018-04-23 08:13

本文选题：推荐系统 + 矩阵因子分解　；参考：《计算机应用研究》2017年02期

【摘要】：在推荐系统中,因评分尺度差异而造成的偏差问题一直影响着协同过滤算法的预测准确性。其中针对矩阵因子分解算法中的偏差问题,提出一种基于高阶偏差的因子分解机算法。该算法首先按照评分偏差的现实特征对用户和项目进行划分,再将偏差类别作为辅助特征集成到因子分解机中,实现了评分预测中不同偏差用户、项目的高阶交互。在MovieLens数据集上的实验结果表明,相比传统矩阵因子分解算法,提出的算法具有更低的预测误差,体现了其更好的推荐性能。
[Abstract]:In the recommendation system, the deviation caused by the difference of scoring scale has always affected the prediction accuracy of collaborative filtering algorithm. In order to solve the deviation problem in matrix factorization algorithm, a factorizer algorithm based on higher order deviation is proposed. In this algorithm, users and items are divided according to the actual features of scoring bias, and then the deviation categories are integrated into the factoring machine as auxiliary features to realize the high-order interaction between users and items with different deviations in scoring prediction. The experimental results on the MovieLens dataset show that the proposed algorithm has lower prediction error than the traditional matrix factorization algorithm, which reflects its better recommendation performance.
【作者单位】：上海大学管理学院;
【基金】：国家自然科学基金资助项目(11201290,61104042)
【分类号】：TP391.3

【参考文献】