一种基于矩阵分解的电影推荐算法

A movie recommendation algorithm based on matrix decomposition

针对目前的电影推荐算法中,传统的矩阵分解算法对于用户的离散型评分数据集的数据利用率不高的问题,提出基于二项分布的矩阵分解算法的模型,在假定用户的评分数据是服从二项分布的前提下,利用最大后验估计学习得出损失函数,将用户的兴趣度作为影响因子,加入项目之间的邻域影响,其后利用随机梯度下降法针对问题求解。通过在MovieLens数据集上与传统的矩阵分解算法的对比实验,结果表明,提出的算法可以有效的提高推荐精度,表现出良好的稳定性。

For the current movie recommendation algorithm, the traditional matrix factorization algorithm score for the user＇s discrete data set is not high data availability problem, matrix decomposition algorithm based on the binomial model, on the assumption that the user data is to obey ratings under the premise of the binomial, using the maximum a posteriori estimation loss function study results, the user＇s interest degree as the impact factor, join neighborhood impact between projects,followed by the use of stochastic gradient descent method for problem solving. By the Movie Lens datasets with conventional matrix factorization algorithm comparison test results show that the proposed algorithm can effectively improve the recommendation accuracy, showing good stability.