基于中位数的用户信誉度排名算法

针对推荐系统易受Spammer攻击的影响,从而导致对象的实际得分不准确的问题,提出基于中位数的用户信誉度排名算法。通过衡量用户信誉度调整用户打分权重,根据中位数具有不易受极端打分影响的特性,选取用户打分与对象得分差距的中位数作为降低用户信誉度的标准,不断迭代调整用户信誉度以及最终得分直至收敛。在多个真实数据集上的运行结果证明,相比现有排名算法,该算法具有更合理的信誉度分布和更高的排名结果准确度,通过该算法预处理后的数据集在SVD++上运行可以得到更低的均方根误差。

User Reputation Ranking Algorithm Based on Median

For the problem that the recommendation system is vulnerable to the impact of Spammer attack, which leads to the inaccuracy of the final item rating, this paper proposes a user reputation ranking algorithm based on median. The algorithm readjusts the weight of user＇s rating by measuring user＇s reputation. On the other hand, according to the median, it has the property of less susceptible to the effects of extreme rating, the algorithm selects the median from the distances between user rank and object rank as the criterion to decrease user reputation, then iterates until convergence to adjust the user reputation and final rating. Operation result of multiple real data sets shows that the algorithm obtains a more reasonable reputation distribution and a higher accuracy, and after preprocessing by this algorithm, the rating data can get a better Root Mean Square Error（RMSE） value on SVD＋＋.