一种改进专家信任的协同过滤推荐算法

Improved expert trust collaborative filtering recommendation algorithm

针对传统基于用户的协同过滤算法较少考虑信任对象所处环境的实时变化,提出一种结合社交网络的专家信任推荐算法。为更好地量化对象之间的信任度,首先利用专家的评价可信度、活跃度、评价偏差度等量化因子计算得到专家的信任值;其次在评分形成的过程中与近邻算法相融合,明确用户与"专家"和"近邻"的偏好,当可选专家人数小于预先设定的阈值时,利用协调因子动态调整近邻算法与改进专家算法的权重,以便获得更加客观的项目评分。最终实验结果表明,在不同大小的Movie Lens数据集上相比于传统的算法,提出的推荐算法在实时推荐预测准确度方面有显著提高。

To solve the tradition collaborative filtering recommendation based on the user object with little consideration of trust real-time change, this paper proposed an expert＇ s algorithm combined with social network trust to make good quantitative trust among objects. Firstly, the algorithm computed evaluation credibility, the active degree and the deviation degree to acquire the expert＇ s trust value. Then, it integrated the scores and neighbor algorithm, and determined the preference of the users and the experts, the users and the neighbor. When the numbers of the selected experts were less than the threshold, it used coordination factor to adjust dynamically the weight between the neighbor algorithm and the expert＇ s algorithm to get more objective score. Experimental results show that proposed algorithm achieves better result on the accuracy of recommendation.