融合时间衰减与偏好波动的协同偏好获取方法

Preference prediction method based on time attenuation and preference fluctuation

针对现有的推荐系统多采用近邻用户的偏好行为来预测当前用户的偏好,而不考虑用户的偏好会随着时间的变化而改变,影响了推荐准确率的问题,提出了一种基于时间衰减与偏好波动的协同偏好获取方法。首先,基于时间因素、用户历史偏好等获取偏好衰减增量与衰减速度,并据此生成衰减函数,使用衰减函数对用户历史行为数据进行衰减修正;其次,基于用户的历史偏好分布获取其偏好波动幅度;最后,将衰减函数与偏好波动幅度分别加入到最近邻获取与偏好获取流程,协同为用户生成推荐列表。在大规模真实数据集上的实验结果表明,所提出的方法与基于属性评分分布的协同过滤(RDCF)与最优Top-N的协同过滤(OTCF)相比,平均绝对误差(MAE)值分别降低了近6.42%和7.73%。实验结果表明所提方法能够提高推荐准确度,提升推荐质量。

The existing recommender systems often use the nearest neighbors＇ preference behavior to predict current users＇ preference, and their recommendation accuracy are influenced by the lack of consideration that users＇ preference would change over time. To solve this problem, a cooperative preference prediction method based on time attenuation and preference fluctuation was proposed. First, attenuation increment and attenuation speed were obtained based on time and historical preference, and the attenuation function was generated by attenuation increment and attenuation speed to modify users＇ historical preference behavior. Then the distribution of historical preference was used to compute the preference fluctuation range. Finally, the recommender list was generated for user by applying the attenuation function and preference fluctuation range into the acquisition of nearest neighbors and the preference acquisition process. The experimental results on real data set show that, compared with the Collaborative Filtering based on Rating Distribution（ RDCF） and Optimizing Top-N Collaborative Filtering（ OTCF）, the average Mean Absolute Error（ MAE） of the proposed method is decreased by about 6. 42% and 7. 73%respectively. It also shows that the proposed method can achieve higher recommendation accuracy and better recommendation quality.