一种基于协同矩阵分解的用户冷启动推荐算法

发布时间：2018-02-12 14:03

本文关键词： 推荐系统位置服务概率矩阵分解冷启动问题约束　出处：《计算机研究与发展》2017年08期 　论文类型：期刊论文

【摘要】：位置服务作为一种信息共享平台,在方便人们交流和共享信息的同时,也因为用户数量的不断增加,而面临着严重的信息过载问题.如何利用推荐技术对信息进行过滤和筛选,帮助用户在位置服务中发现有价值的信息成为近年来研究的热点.但目前已有的推荐算法,在只有消费记录这种隐性数据情况下,针对用户较少活动区域或新用户的推荐效率较低,无法最大化挖掘隐性数据所带的信息.针对以上问题,结合位置服务平台的特点,针对用户冷启动问题,提出了一种结合协同概率矩阵分解与迭代决策树(gradient boosting decision tree,GBDT)的推荐算法.该方法首先使用多层协同概率矩阵分解在多个维度上得到用户潜在特征,然后使用GBDT学习算法对特征和标签进行训练得到用户对项目的偏好,最后使用考虑约束问题的top-N推荐产生推荐列表.在真实数据集上的实验结果表明,与目前较为流行的方法相比,提出的方法能在准确率、F1值上取得较好的结果,能更好地缓解位置服务中的冷启动问题.
[Abstract]:As a kind of information sharing platform, location service not only facilitates people to exchange and share information, but also faces serious information overload problem because of the increasing number of users. Helping users find valuable information in location services has become a hot topic in recent years. The recommendation efficiency for the users with less active areas or new users is low, which can not maximize the information of mining the hidden data. In view of the above problems, combined with the characteristics of the location service platform, the cold start problem of users can be solved. In this paper, a recommendation algorithm combining cooperative probability matrix decomposition with iterative decision tree gradient boosting decision treedoff (GB DTT) is proposed. In this method, the potential characteristics of users are obtained by using multi-layer cooperative probability matrix decomposition on multiple dimensions. Then the GBDT learning algorithm is used to train the feature and label to get the user's preference for the item. Finally, the top-N recommendation with constraint problem is used to generate the recommendation list. The experimental results on the real data set show that, Compared with the current popular methods, the proposed method can obtain better results on accuracy and F1 value, and can better alleviate the cold start problem in location services.
【作者单位】：山东师范大学信息科学与工程学院;山东师范大学管理科学与工程学院;山东大学计算机科学与技术学院;
【基金】：国家自然科学基金项目(61602282,61602284) 中国博士后科学基金项目(2016M602181) 国家社会科学基金项目(14BTQ049)~~
【分类号】：TP391.3

【参考文献】