引入偏置的矩阵分解推荐算法研究

Bias based matrix factorization recommender techniques

由于矩阵分解良好的可扩展性和较高的预测精度,在推荐算法的应用中有出色的表现。在基础的矩阵分解模型上先后引入全局偏置和时间偏置,以提高预测准确度和推荐质量。以个性化推荐系统为对象,在Movie Lens数据集上的实验表明,所设计的方法提高了算法的预测精度。

Matrix factorization which is widely accepted in recent recommendation prize has become popular by combining good scalability with predictive accuracy and much flexibility.This paper focused on modeling bias and time effects in matrix factorization.And it tested these models on MovieLens dataset.Experiment results show that prediction accuracy in matrix factorization can be improved significantly by using bias and time information.