融合相似用户和信任关系的动态反馈协同过滤推荐算法

为了克服推荐算法的静态性缺点,提出融合相似用户和信任关系的动态反馈协同过滤推荐算法.该算法用动态因子融合相似用户和信任关系,动态因子初始取随机数,根据用户反馈和系统预测的误差建立正负反馈机制.按照反馈类型,选择增值或衰减函数适当调整动态因子,以便系统更好预测用户评分.在真实数据集Epinions上的实验表明,采用正负反馈的动态融合算法,不仅克服了静态性缺点,而且较基于相似用户或者信任关系的推荐进一步提高了推荐准确率.

Collaborative filtering recommendation algorithm of dynamic feedback combining similar users with trust relationship

The recommendation algorithm fusing similar users and trust relationship can handle the cold boot, whereas exists the static flaw. In order to solve the problem, the collaborative filtering recommendation algorithm of dynamic feedback combining similar users with trust relationship is proposed. The algorithm integrates similar users with trust relationship using dynamic factors, which are randomly initialized. The positive and negative feedback mechanism is established according to the error between user feedback and system prediction. In the light of the type of feedback, the value added or attenuation function is selected to properly adjust dynamic factors so that the system better predicts the user"s score. Experiments using the real data set Epinions show that the dynamic fusion algorithm with the positive and negative feedback further improves the recommendation accuracy than that based on similar users or trust relationship.