融合标签相似度的 k 近邻 Slope One 算法

Slope One 协同过滤算法被广泛应用于个性化推荐系统中。标签是一种描述项目特性的重要形式，针对Slope One 算法推荐精度不足的问题，将标签信息融合到 Slope One 算法当中。同时参考 k 近邻算法思想，选取阈值过滤后的 k 近邻项目参与平均评分偏差计算，提高计算效率的同时增加预测精度。使用评分相似度和标签相似度作为权重修正线性回归模型。通过线性加权融合预测结果，进一步提升推荐质量。将算法应用于 MovieLens 数据集，与传统加权 Slope One 算法相比，平均绝对偏差下降4．8％，召回率和准确率分别提高32．1％和26．3％。

K-nearest neighbor hybrid Slope One algorithm combined with tag similarity

Slope One Collaborative Filtering algorithm is widely used in personalized recommendation system. Label is an important form to describe the characteristics of the items. To overcome its deficiency in rating prediction accuracy, this paper proposes a new hybrid algorithm combined with tag information. With reference to the k-nearest neighbor Collaborative Filtering algorithm, we select neighbors of the target item to participate in the calculation of the average rating deviation,which ensures computational efficiency and improves the prediction accuracy. The algorithm defines rating similarities and tag similarities as weight to revise the linear regression model. To achieve further improvement of the recommendation quality, the algorithm adopts a linear weighted fusion method to combine the results. Experimental results on the Movielens data sets indicated that, compared with the traditional weighted Slope One algorithm, mean average absolute error declined 4.8% , while recall rate and precision rate respectively increased 32.1% and 26.3% .