基于项目候选集的协同过滤算法

针对协同推荐技术中的数据稀疏性等问题，提出了一种新颖的协同过滤算法。该算法在用户相似的协同过滤基础之上，引入用户之间的非对称影响度和支持度，提出项目候选集的概念。并且，该算法利用项目信息熵对未评分项目集合进行二次修正，得到候选项目集合，进而提高算法的准确率。基于MovieLens和netflix数据集的实验表明，相比目前多个比较流行的算法，该算法在推荐结果的准确率、召回率和F1值上都具有较大的优势，有效地降低了用户评分稀疏性问题带来的负面影响，显著提高了推荐系统的推荐质量。

Collaborative filtering recommendation algorithm based on candidate items

To alleviate the impact of data sparsity in traditional collaborative filtering algorithm, the paper proposed a collaborative filtering recommendation algorithm based on candidate items. The asymmetric impact and support level between users were first introduced into the user-based collaborative filtering algorithm, and then a new concept of measuring candidate items was presented by applying the item entropy-based methods to modify the preliminary items which did''t rate. Experimental results on MovieLens dataset and netflix dataset showed that the proposed algorithm outperformed many current similar algorithms in measurements of accuracy, recall and F1 values.