基于信任关系的潜在好友推荐方法

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。

The method of latent friend recommendation based on the trust relations

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.