社会网络中的推荐系统通过捕捉用户在社会网络中体现的兴趣爱好，使之获得个性化的信息服务，是众多社会网络应用的核心支撑技术之一。已有的推荐算法大多假设社会网络中只有一种用户关系，但事实上社会网络及其扩展应用中存在着多种类型的用户关系，如人际关系包括朋友、同事等，互动关系可包括共同购买同一商品、在线通讯等。不同的关系对于特定的推荐任务所起的作用是大不一样的。充分利用多种用户关系对实现高质量的社会网络推荐系统具有十分重要的作用。为了将这些关系运用到推荐系统中，本研究首先为社会网络中的多种用户关系进行建模；为体现不同用户关系对推荐任务的不同作用，设计具体的关系分析方法；接着，在如何使用多种用户关系提高推荐系统的性能方面，本研究分别基于协作过滤、机器学习和数据挖掘技术为多关系社会网络设计若干推荐算法并进行评估；最后使用真实在线社会网络数据对所述模型算法进行验证，分析多社会关系推荐机制的准确性和健壮性。

Recommendation systems in social network can personalize the information service for users through capturing their interests and hobbies embodied in the social network, which is one of the core supporting technologies of many social network applications. Most of the existing methods on social recommendation assume that there is only one kind of user relation in the social network. However, in reality, there exist various kinds of users' relations in social network and its extended applications, such as interpersonal relationships include friendship relations, colleague relations, etc., and interactive relationships include purchasing the same product, online messaging, etc. Moreover, each kind of relation may play a distinct role in a particular recommendation task. The fully exploitation of these multiple relations in social network will play a very important role in the realization of high-quality recommendation system. In order to combine a rich variety of relations of users in social network into recommendation system, our research firstly establishes the models of these various kinds of relations. And in order to distinct different importance of different relations with respect to a particular recommendation task, we introduce several relation analysis methods. To enhance the performance of recommendation system by utilizing the multiple relations, our research also address several social recommendation algorithms based on collaborative filtering, machine learning and data mining respectively. At last we verify our proposed models and algorithms in real online social network and analysis the accuracy and robustness of our multi-relational social network recommendation mechanism.