现今，隐语义模型是在音乐推荐系统中被广泛运用的一种模型，通过矩阵 分解将“用户-评分”矩阵降维成“用户-特征”和“物品-特征”矩阵，利用 损失函数最小化训练习得该模型。隐语义模型极大的降低了常见的基于邻域 的协同过滤模型的计算复杂度。在本篇论文中将进一步优化该模型。考虑到用 户物品个体特殊偏好属性的影响，加入物品和用户偏置项；加入“用户-特征” 中用户对特征的邻域因素，优化“用户-特征”矩阵；在偏置项中用分桶、构 造关于时间 t 的 drif 函数，在模型中融合时间因素。最后利用 Amazon 的数 字音乐数据库对隐语义模型参数和性能进行分析。

Nowadays, Latent Factor Model (LFM) is widely used in music recommendation systems. Through matrix decomposition, the "user-rating" matrix is reduced into "user-feature" and "item-feature" matrices. Minimizing the loss function is to train the model. The Latent Factor Model model greatly reduces the computational complexity of the common neighborhood-based Collaborative Filtering Model. In this paper, the model will be further optimized. Taking into account the influence of the particular preference attributes of the user’s items, add the items and user bias items; add the neighborhood factors of the user to the feature in the "user-feature" to optimize the "user-feature" matrix; use binning in the bias item , Construct the drif function about time t, and integrate the time factor in the model. Finally, use Amazon's digital music database to analyze the LFM parameters and performance.