面向社会网络用户需求的推荐系统研究

At present, users and contents over the social network are growing dramatically. Internet users retrieve, communicate, and share information in the social network. The development of social media helps the spreading of informaiton. At the same time, however, users have to analyze the credibility of the massive communicated information, mostly generated by users, before making information consumption. Based on users' various information requirements, we proposed novel approachs for developing distributed collaborative filtering algorithms; building social media user preferences to characterize the user behavior over the social media; designing trust model to discover the trust relationships between social network users. More specifically, we intent to pursue researches on the following aspects: Distributed Adaptive Collaborative Filtering (CF): To overcome the limitation of performance of the traditional CF for dealing with massive user rating data, we proposed a divide-and-conqure approach for the item-based recommender system. In this study, we advocate that the ranking performance is more important than the average predicted ratings for recommender system. We also provide an incremental learning, such that parameters can be easily updated as new ratings becomes available. Content Recommendation based on the User Preference Model: This refers to characterizing the relations between user and contents. This model can map user with topic interests and similar behavioral patterns. Such information may draw users of common interest into the groups to enhance their experience in interacting with other users thereby further engaging the users. User Recommendation based on Trust Prediction: When determining the trustworthiness of social members, traditional approaches merely consider the similarity between users. However, the propagation of trusts of users in the social network also play an important role for precisely model the trust network. In this work, we exploit factor graph and belief propagation within the social network to globally analyze the trust relationships between users, and then recommend reliable users and friends.

随着网络技术的日益成熟，人们逐渐依赖互联网获取信息，交流沟通。社会网络技术的成长使得社会网络中用户间的交流越来越密切，社会网络逐渐成为用户进行信息获取和交互的重要平台。社会化媒体在方便用户交流的同时，也导致用户面对大量信息而无法判断其可靠性。传统推荐系统在处理大规模社会网络数据时往往会遇到计算效率和算法性能等瓶颈，社会网络中数据的多样化也需要提供新的用户建模方法和信息推荐模型满足社会网络用户需求。本课题面向社会网络用户对社会产品评分，社会媒体内容和社会成员关系三个方面的推荐需求展开研究,拟提出分布式自适应的协同过滤计算方法克服传统推荐系统处理大规模数据的性能瓶颈；采用图模型对社会网络中的传播内容与用户间关系进行建模，分析用户的潜在偏好并基于生成者和接受者的偏好分布生成媒体内容推荐；通过对社会网络中成员信任关系使用数学方法建模,建立用户信任关系传递模型并给出可信社会成员推荐。

随着社交媒体用户数量的指数级增加，社会网络规模不断扩大，给传统推荐系统处理大规模数据带来了巨大挑战。因此，设计高效精准的推荐方法对于突破传统推荐系统的性能瓶颈，辅助社会网络用户快速获取高质量的信息有着重要的作用。同时，为用户提供更安全、更快捷、更准确的信息推荐，满足互联网中社会网络用户的个性化需求，创造社会价值和商业价值。.本项目在充分调研目前主流推荐系统的基础上，面向社会网络用户需求，针对社会网络和社会媒体中非结构化信息传播的几何级扩散所导致信息过载和内容可信的问题，开展的推荐系统研究，为社会网络用户在海量信息中找到可靠的，高可用性的数据提供帮助。在研究内容方面，本项目面向社会网络用户对于产品资源、媒体内容和可信社会成员等方面的个性化需求，解决现有推荐系统在社会网络环境下协同过滤算法、用户偏好模型、社会成员信任关系分析及预测和个性化推荐等方面存在的问题。.在本项目执行期间，项目组提出了一种新的分布式自适应协同过滤算法；结合概率图模型与社会网络分析技术，分析用户对不同内容的接受模式，构建了一系列的个性化社会媒体内容推荐模型；抽象影响用户信任关系的因素，并建立社会网络用户的信任模型，模型在准确率和效率上都要比传统方法具有优势。同时，项目执行期间部分成果与中国民航信息股份有限公司合作，协助航空公司和代理商实现基于旅客消费特征和兴趣偏好的多种产品的交叉销售和动态打包推荐，实现了部分研究成果的应用转换。