一种基于用户兴趣转移挖掘的流式推荐算法  

A STREAMING RECOMMENDATION ALGORITHM BASED ON USER INTEREST DRIFT MINING

作　　者：陈建宗 刘永坚[1] 解庆 唐伶俐[1] Chen Jianzong;Liu Yongjian;Xie Qing;Tang Lingli(School of Computer Science and Technology,Wuhan University of Technology,Wuhan 430070,Hubei,China)

机构地区：[1]武汉理工大学计算机科学与技术学院

出　　处：《计算机应用与软件》2020年第1期59-65,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(61602353);湖北省自然科学基金项目(2017CFB505)

摘　　要：推荐系统是当下解决信息超载的有效方法,但由于用户兴趣转移现象的存在,传统推荐系统在时间跨度较长的应用场景下表现并不理想。为了解决该问题,提出一种基于用户兴趣转移挖掘的流式推荐算法。根据资源的种类信息构建资源特征向量,采取增量更新方法,根据流数据实时更新模型参数,避免了传统增量矩阵分解模型中的拟合残差扩大问题。模型结合提出的两种新型遗忘机制,能够有效区分用户历史数据中的临时偏好与长期偏好,从而在遗忘用户过时数据的同时,保留用户的长期偏好。在电影推荐数据集中进行实验,证明了该算法的有效性。Recommendation system is an effective method to solve the problem of information overload.However,due to the existence of user interest drift,the performance of conventional recommendation system is unsatisfactory in the applications with a long time span.To solve this problem,this paper proposes a streaming recommendation algorithm based on user interest drift mining.To avoid the fitting error occurred in the conventional incremental matrix factorization,our model constructed resource feature vectors based on the information of resource types,and updated model parameters in real time according to stream data by incremental updating method.Two kinds of novel forgetting mechanisms were embedded in the proposed model,which could effectively distinguish temporary preferences from long-term preferences in user historical data,so as to preserve users long-term preferences while forgetting users outdated data.Experiments on the movie recommendation dataset demonstrate the effectiveness of the proposed algorithm.

关 键 词：推荐系统 兴趣转移 流数据挖掘 增量矩阵分解 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]