融合矩阵分解与距离度量学习的社会化推荐算法

为解决传统推荐系统中存在的冷启动难题,基于距离反映偏好的假设提出了一种融合矩阵分解与距离度量学习的社会化推荐算法。该算法同时对样本和距离度量进行训练,在满足距离约束的前提下更新距离度量和用户与项目的坐标,并将用户与项目嵌入到统一的低维空间,利用用户与项目之间的距离生成推荐结果。基于豆瓣和Epi-nions数据集的对比实验结果验证了该方法可有效提高推荐系统的可解释性和精确度,明显优于基于矩阵分解的推荐方法。研究结果表明,所提方法缓解了传统推荐系统中存在的冷启动问题,为推荐系统的研究提供了另一种可供参考的研究思路。

Social Recommendation Method Integrating Matrix Factorization and Distance Metric Learning

In order to solve the dilemma called cold start in traditional recommender systems,a novel social recommendation method integrating matrix factorization and distance metric learning was proposed based on the assumption that distance reflects likability.The algorithm trains the samples and distance metric,at the same time,the distance metric and the coordinates of users and items are updated to meet the constraints of distance.Finally,users and items are embedded into an united low dimensional space,and the distance between users and items is used to generate recommendation results.The experimental results on Douban and Epinions datasets show that the proposed method can effectively improve both interpretability and accuracy of recommender systems and is superior to recommendation methods based on matrix factorization.Research results indicate that the proposed method mitigates the cold start dilemma in traditionalrecommender systems,and it provides another research idea for recommender systems.