基于信任关系的潜在好友推荐方法  

The method of latent friend recommendation based on the trust relations

作　　者：黄亮[1] 杜永萍[1] 

机构地区：[1]北京工业大学计算机学院,北京100124

出　　处：《山东大学学报（理学版）》2013年第11期73-79,共7页Journal of Shandong University(Natural Science)

基　　金：国家自然科学基金资助项目(60803086);国家科技支撑计划子课题项目(2013BAH21B02-01);北京市自然科学基金资助项目(4123091);北京市属高等学校人才强教深化计划"中青年骨干人才培养计划"项目(PHR20110815)

摘　　要：如何有效地帮助用户挖掘平台潜在好友成为电子商务中一项非常重要的服务需求。提出了一种综合考虑用户间兴趣因素和信任因素的好友推荐方法，设计并构建了一个包括用户声望信任和局部信任的混合信任网络，将网络中信任评价度与协同过滤中兴趣评分相似度进行组合来衡量用户间好友相似关联，以实现好友推荐。在Epinions数据集上以准确率、召回率和F值作为实验评价指标，对所提方法进行验证，相比其他同类应用准确率在10％-15％、召回率在10％～20％的性能，本文方法的准确率和召回率的最佳性能分别达到22．47％和21．15％，实验证明本文方法有效提高了推荐性能。E-commerce has greatly changed people＇s daily activity and consumption behavior. Mining the potential plat- form friends effectively has become an important demand for services in the e-commerce. We present an approach for the friend recommendation by the consideration of user interest factors and trust factors. The mixed trust network is de- signed and built and it contains the authority value and the trust information between the users. In order to achieve per- sonalized friend recommendation, the trust evaluation value and the similarity value based on the interest are combined to measure the association between the users. The experiment on the Epinions dataset is carried and the precision, recall and F-value are used as the evaluation metric. Compared to other system of precision 10%-15% and recall 10%-20%, the best performance of precision 22.47% and recall 21.15%. The results show that the proposed method effectively improves the recommended performance.

关 键 词：声望信任 信任计算 好友推荐 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]