Truser:一种基于可信用户的服务推荐方法  

Truser: An Approach to Service Recommendation Based on Trusted Users

作　　者：何鹏 吴浩 曾诚 马于涛[3] HE Peng;WU Hao;ZENG Cheng;MA Yu-Tao(School of Computer and Information Engineering, Hubei University, Wuhan 430062;School of Computer Science, Wuhan University, Wuhan 430072;Engineering Technology Research Center for Education Informatization, Hubei Province, Wuhan 430062)

机构地区：[1]湖北大学计算机与信息工程学院,武汉430062 [2]湖北省教育信息化工程技术研究中心,武汉430062 [3]武汉大学计算机学院,武汉430072

出　　处：《计算机学报》2019年第4期851-863,共13页Chinese Journal of Computers

基　　金：国家重点研发计划(2017YFB1400602);国家自然科学基金(61572371);湖北省技术创新重大专项(2018ACA13);湖北省教育厅青年人才项目计划(Q20171008)资助~~

摘　　要：在服务推荐过程中,为排除不可信用户信息带来的干扰,确保推荐结果的精准性,该文从用户聚类的角度,通过两阶段的ISODATA聚类,将离群用户视为不可信用户进行过滤,再基于得到的可信用户提出一种改进的服务推荐方法.最后,在两个公开数据集Last.FM和Delicious上进行了实证分析.结果表明,该文所提方法在两个数据集上的推荐精度相较于已有基准方法分别提高16.1%和4.5%,且发现当第一阶段聚类的预期聚类中心为6时,推荐效果最好;同时,在推荐过程中为目标用户返回Top-5个可信用户,且向其推荐这5个用户中至少有70%的人关注过的服务最为适宜.因此,围绕可信用户的数据进行推荐,能有效地提高服务推荐的质量.As a very important topic in the field of service computing, service recommendation has been paid much attention by researchers. To improve the service recommendation quality and ensure user’s experience of services, various methods have been proposed successively. However, most existing methods mainly focus on how to improve the accuracy of recommendation models by introducing richer information or advanced modeling techniques, and there is a general lack of discussion on users’ trustworthiness from the data quality point of view. In practice, untrusted users are common for a variety of reasons, such as fake comments. Data generated by untrusted users is often misleading and unhelpful for recommendations. Therefore, it is very necessary to eliminate noise from untrusted users before service recommendation;otherwise, the quality of recommendations will always be affected regardless of how the model is optimized. To achieve this, we identify untrusted users based on their abnormal labeling behavior compared to the public and attempt to filter out these outliers before service recommendation, from the perspective of a two-stage ISODATA (Iterative Self-Organizing Data Analysis Technique Algorithm) clustering, and then propose a novel approach to service recommendation based on the resulting trusted users, named as Truser. Compared with the conventional K -means clustering algorithm, ISODATA is more flexible and increases the operation of “merging” and “splitting” to adjust the clustering number. In other words, the clustering center K value of ISODATA can be adjusted dynamically according to the actual situation. In this paper, we first perform ISODATA clustering for the concerned users of each service to label its candidate untrusted users. Then we get the number of times each user is marked as candidate untrusted, and the number of times any two users are clustered into the same group. According to the information obtained above, we further utilize ISODATA to cluster users, so as to filter out untruste

关 键 词：ISODATA聚类 协同过滤 服务推荐 服务计算 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]