基于标签分类的协同过滤推荐算法

传统的协同过滤根据用户的行为去预测可能喜欢的产品,是当前应用最广泛的推荐算法之一。但随着用户规模的急剧扩大,有价值的信息占比较少,存在稀疏性等问题,导致推荐质量不高。针对这一问题,提出了一种基于标签分类的协同过滤推荐算法。将不完整的数据样本根据标签进行分类,使分解的矩阵依赖于类,随后使用迭代投影寻踪的方法计算类依赖矩阵的线性组合及其对应的权重。开放数据集实验表明,该方法在保持一定分类准确率的前提下,平均降低了35.23%的插补误差,优于传统协同过滤推荐算法。

Collaborative filtering recommendation algorithm based on label classification

Traditional collaborative filtering is one of the most widely used recommendation algorithms based on the user behavior. However, with the rapid expansion of the user scale, there are fewer valuable information so that it leads to bad recommendation quality because of matrix sparsity. To solve this problem, this paper proposed a collaborative filtering recommendation algorithm based on label classification. Incomplete data samples were categorized according to the labels so that the decomposed matrix could depend on the class. Then the linear combination of class-dependent matrices and its corresponding weights were calculated by using iterative projection pursuit. The experiments of open datasets show that the proposed method reduces the average interpolation error by 35.23% while maintaining certain classification accuracy. This method is better than the traditional collaborative filtering recommendation algorithm.