基于社交信息和物品曝光度的矩阵分解推荐

Matrix factorization recommendation based on social information and item exposure

为了解决隐式反馈推荐中的数据稀疏性和未观测值二义性,提出基于社交信息和物品曝光度的概率矩阵分解推荐算法.该算法通过对用户-用户社交矩阵进行矩阵分解来约束用户偏好潜在因子,一定程度上缓解了数据稀疏性问题;将物品曝光度作为观测值的条件,结合物品本身的流行度和用户的社交信息,对物品曝光度进行建模,解决未观测值的二义性.在Lastfm公开数据集上开展多个层次的实验和分析.结果表明,与已有的隐式推荐算法相比,在召回率、平均准确率(MAP)和归一化折损累计增益(NDCG) 3个评价标准上都有一定程度的提高.

A probabilistic matrix factorization model was proposed based on social information and item exposure in order to solve the problem of data sparsity and ambiguity of unobserved data in implicit feedback recommendation. The model constrained users’ potential preference factor by decomposing the user-user social matrix to alleviate data sparseness to some extent. The item exposure was taken as the condition of the observation value, and the item exposure was modeled based on the popularity of the item itself and the social information of the user in order to resolve the ambiguity of the unobserved data. Multiple levels experiments and analysis were conducted on Lastfm dataset. Results show that the proposed model can improve the performance on metrics of recall, mean average precision (MAP) and normalized discounted cumulative gain (NDCG) compared to state-of-art implicit feedback algorithms.