分步填充缓解数据稀疏性的协同过滤算法

为解决传统协同过滤算法中用户评分数据稀疏性所带来的用户最近邻寻找不准确问题,提出了一种结合条件概率和传统协同过滤算法的非固定k近邻算法。该算法在基于分步填充评分矩阵的思想上,第一步只接受相似度和共同评分项目数量达到阈值的邻居用户作为目标用户邻居,然后计算并填充未评分项目,第二步使用第一阶段部分填充后的矩阵计算剩余未评分项目的评分。实验表明,该算法寻找的邻居用户更准确,能明显地缓解数据稀疏问题,提高评分预测准确性。

Collaborative filtering algorithm based on two-step filling foralleviating data sparsity

To overcome the data sparsity of traditional collaborative filtering algorithm which can cause inaccuracy during finding the nearest-neighbors, the paper came up with a new method combining conditional probability with traditional collaborative algorithm whose neighbors was not always k. The algorithm' core was that the last data matrix was calculated by two-step filling. The first step, it accepted the users whose similarity and the number of both- rated items met the standard as the target user's neighbors and then calculated the value and filled the unrated- items. The second step would fill the left unrated- items relying on the data matrix filled by the first step. Experimental results show that this algorithm can find reliable neighbors, alleviate the data sparsity and achieve better prediction accuracy obviously.