融合社交信任的多属性元路径好友推荐方法

Friend Recommendation Method Based on Multi-attribute Meta-path with Social Trust

随着社交网络的日益繁荣,好友推荐成为了社交网络提供的重要服务之一,也成了推荐方法研究的热点.现有的好友推荐方法大多基于用户的兴趣偏好进行推荐,未考虑在线社交的异构网络特征,不能有效抵御恶意用户的托攻击,且计算量相对较大.提出了一种融合社交信任的多属性元路径好友推荐方法,该方法分析了在线社交的异构网络特征,使用元路径理论对好友推荐的异构网络进行抽象和形式化,提出了用户本地信任网络模型(ULTNM).该信任模型综合考虑元路径上的用户社交圈子相似度和统计兴趣偏好相似度等多种属性特征,构建了目标用户的本地信任网络,并基于Ford Fulkerson算法对信任网络进行搜索和排序,进行目标用户的候选好友推荐.基于真实数据集的对比实验表明,该推荐方法有更好的推荐准确率和召回率,能更好的抵御恶意用户的托攻击行为,且运行效率具有一定的优势.

With the grow ing popularity of social netw orks,friend recommendation becomes one of important service supplied by social netw orks,and it is also the hot research filed of recommendation studies.The existing friend recommendation methods mostly based on the interest preference of users,cost computational resource heavily,and w eakly w hile under the Shilling attacks of malicious users.Aims to solve this problem,this paper propose a friend recommendation method based on multi-attribute meta-path w ith social trust.At first,w e use meta-path theory to formalize the heterogeneous netw ork of friend recommendation,then propose the ULTNM(User Local Trust Netw ork M odel)trust model based on the social circle similarity and the statistic interest similarity among users.After that,w e use the ULTNM to build the local trust netw ork of the target user,then use Ford Fulkerson algorithm to search the local trust netw ork to find friend candidates of the target user,and sort these friend candidates finally.The comparing experimental results w hich based on the Epinions data sets show that our recommendation method has better precisions and recalls,and it can count against the Shilling attacks better w hile comparing to state-of-art friend recommendation methods,and it also costs less computational resource.