缺失数据建模的改进型ALS在线推荐算法

Improved ALS Online Recommendation Algorithm with Missing Data Modeling

在隐式反馈中存在数据噪声并缺乏负反馈,使用矩阵分解方法训练推荐模型时特征提取不明确且推荐结果有偏斜。为此,提出一种缺失数据建模的改进型交替最小二乘(ALS)矩阵分解在线推荐算法。使用近邻信息为用户选择正样本,同时根据物品流行度对缺失数据中的负样本进行建模,并将基于元素的改进型ALS算法与在线学习相结合。在MovieLens数据集上的实验结果表明,相对eALS、Rcd算法,该算法能够有效减小数据噪声和矩阵稀疏对矩阵分解推荐算法的影响,从而提高推荐的准确性和效率。

Implicit feedback has data noise and lacks negative feedback. When using matrix decomposition method to train recommendation model,the feature extraction is not clear and the recommendation result is skewed. Aiming at these problems,an improved Alternating Least Square（ ALS） matrix decomposition online recommendation algorithm with missing data modeling is proposed. The neighbor information is used to select positive samples for the user,at the same time,the negative samples in missing data are modeled according to the popularity of articles. Then the improved ALS algorithm based on elements is combined with online learning. Experimental results on MovieLens data sets show that,compared with eALS and Rcd algorithms,the proposed algorithm can effectively reduce the impact of data noise and matrix sparsity on the matrix decomposition recommendation algorithm,thus improving the accuracy and efficiency of the recommendation.