提出了一种最大化参数变化的主动采样方法,可快速捕捉推荐系统中新用户的兴趣偏好.该方法在纯奇异值分解(PureSVD)模型的基础上,选取最大化模型参数变化的样本,然后向新用户查询样本物品的评分.得到的评分用来训练用户的纯奇异值分解模型参数,进而提供推荐列表.基于贪婪法提出了一种快速的近似采样算法,能在可接受的时间内得到采样列表.实验结果证明,在Movielens数据集上,该方法能在Top—N的标准下使用较小的样本,有效地提高了学习新用户偏好的效率.
A parameter-change maximization sampling method is proposed to capture new user' s prefer- ence in recommender system. This method produces an item list that maximizes model parameter change based on pure singular value decomposition (PureSVD). By querying new user with specific item list, the ratings are obtained for training the corresponding user' s parameter in PureSVD model, it performs prediction for new users in return. A greedy approximation algorithm is presented to produce the item list with an acceptable time bound. Experiments show that the method can learn new user' s preference effi- ciently with small sample size under Top-N metrics.