考虑时序性和动态信任的工程经验知识推荐技术  

Engineering Empirical Knowledge Recommendation Mechanism Considering Time Sequence and Dynamic Trust

作　　者：黄颖[1] 蒋祖华[1] 刘璞凌[1] 王海丽[1] 

机构地区：[1]上海交通大学机械与动力工程学院,上海200240

出　　处：《上海交通大学学报》2016年第9期1422-1429,共8页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金项目(70971085;71271133);上海市教育委员会科研创新重点项目(13ZZ012);上海市科委科技创新行动计划(13111104500);上海汽车工业教育基金会项目资助

摘　　要：随着用户和词条数量的增长,"用户-词条"评分矩阵变得极其稀疏,导致基于相似度计算的推荐算法精度降低.提出一种基于用户短期兴趣时序性变化的评分矩阵进行预填充.由于MediaWiki社区用户群之间的信任呈动态变化,定义了追随率反映同一时间窗内具有相似兴趣的用户对知识推荐的参考性.最后设计实验,确定时间窗长度T的最优参数,通过比较CFBDT(Collaborative filtering based on dynarnic trust)算法与3类现有算法的效果,验证其可行性.Users＇ rating matrix of User-Item-Time model suffers from sparsity severely with the increase of users and items. A pre-filling algorithm to pre-process the rating matrix sequentially was developed to a- void the rapid accuracy loss of recommendation system. Moreover, loyalty L was defined to measure the probability of which a target user would be convinced by an experienced user, as trust between users varied dynamically according to their historical interactions. Finally, experiments were conducted to determine the proper length of time window and verify the effectiveness of the proposed algorithm compared with the three previous collaborative filtering recommendation algorithms.

关 键 词：知识推荐 工程经验知识 动态信任 MediaWiki平台 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]