基于改进带偏置概率矩阵分解算法的研究

针对个性化推荐过程中高维稀疏性问题，本文提出一种将奇异值分解技术和带偏置概率矩阵分解相结合的推荐方法。 首先利用SVD算法初始化用户项目潜在因子向量，避免因随机赋值而使得函数陷入局部最优解，接着将用户项目的偏置信息融入到概率矩阵分解算法中，同时为了提升训练速度和推荐精度，通过动量加速的迷你批量梯度下降（mini Batch Gradient Descent，miniBGD）来训练，最后利用分解后的两个低维矩阵对原矩阵中的未知评分进行预测，在三个公开数据集的实验结果表明，本文提出的算法相对于传统的算法能够有效的提高推荐精度，进一步缓解由数据高维稀疏性带来的推荐质量不高的问题。

Research On Improved Bias Probabilistic Matrix Factorization

To solve the difficulty of data high-dimensional sparse problem in personalized recommendation, the paper proposes a new recommendation algorithm based on singular value decomposition and bias probability matrix decomposition. Firstly, It obtain the user-item rating matrix, then the singular value decomposition initialize the potential factor matrix of users and items, and further the paper integrate bias information with probability matrix decomposition algorithms to improve the accuracy of recommendation. Finally, this paper use maximum likelihood to transform the score prediction problem into an optimization problem by stochastic gradient descent. Simulation experiments on the three publicly available datasets show that the proposed algorithm can improve the recommendation accuracy on the three difference datasets, so as to ease the high-dimensional data sparse.