基于用户偏好和动态兴趣的多样性推荐方法  

Diversified Recommendation Method Based on User Preference and Dynamic Interest

作　　者：邓明通 刘学军 李斌 DENG Ming-tong;LIU Xue-jun;LI Bin(College of Computer Science and Technology,Nanjing Tech University,Nanjing 211816,China)

机构地区：[1]南京工业大学计算机科学与技术学院,南京211816

出　　处：《小型微型计算机系统》2018年第9期2029-2034,共6页Journal of Chinese Computer Systems

基　　金：江苏省重点研发计划(社会发展)项目(BE2015697)资助;国家自然科学基金项目(61203072)资助

摘　　要：协同过滤是目前解决信息过载问题的主要方法之一,然而其推荐的多样性不足,且在冷启动场景下推荐效果较差.提出了基于用户偏好和动态兴趣的多样性推荐方法 DRMUD(A Diversified Recommendation Method Based on User Preference and Dynamic Interest).首先通过对用户历史反馈数据分析用户的多样性偏好,得出用户的多样倾向度;然后引入时间衰减函数,动态调整用户的历史评分数据;最后将矩阵分解和项目疲劳函数相结合,并加入多样倾向度调节两者所占比重.当新用户加入系统时,通过网格索引为其产生最信任邻居,新用户缺失的反馈信息由最信任邻居代替.实验结果表明,DRMUD算法有效缓解了用户冷启动问题,并能在保证准确率的前提下提高推荐结果的多样性.Collaborative filtering is one of the main methods to solve the problem of information overload，but its recommended diversity is not enough，and in the cold start scene recommended effect is poor.A DRDUD （A Diversified Recommendation Method Based on User Preference and Dynamic Interest） based on user preference and dynamic interest is proposed.First，the user′s historical feedback data is used to analyze the diversity preferences of the users，and the diversity tendency of the users is obtained.Then，the time decay function is introduced to dynamically adjust the historical score data of the users.Finally，the matrix decomposition and the project fatigue function are combined and Tendency adjustment of the proportion of both.When a new user joins the system，the most trusted neighbor is generated by the grid index，and the new user′s missing feedback is replaced by the most trusted neighbor.The experimental results show that the DRMUD algorithm can effectively alleviate the cold start problem of the user and improve the diversity of the recommended results under the premise of ensuring the accuracy rate.

关 键 词：冷启动 信任邻居 动态兴趣 多样倾向度 推荐多样性 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]