基于高阶偏差的因子分解机推荐算法

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，本文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，本文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。

High-Order Biased Factorization Machine Recommender Algorithm

In recommender system, bias problem caused by different rating scales has always effected the predict precision of collaborative filtering. Concerning this bias problem of matrix factorization, this paper proposed a high-order biased factorization machine recommender algorithm. Firstly, it grouped users and items by their rating bias feature from real world, then integrated them into the factorization machine, which provided the high-order interactions between the different biased users and items. The experimental results on Movielens datasets demonstrate that the proposed algorithm has lower prediction error than other traditional matrix factorization algorithms, which shows its better recommender performance.