Research On Friend Recommendation Algorithms On Social Networking Services

Online social networking services(or simply social networking services) have been growing exponentially since they emerged just a few years ago. Social networking services have become one kind of Web 2.0 services that create more and more impact on the Internet users. Social networking services have considerably changed how people communicate, think, study, and live, and how organizations do their businesses around the world. Friend recommendation services for online users on social networking web sites can help grow online social networks, improve users’ satisfactions and propagation of knowledge, and enrich the study of social networks.This thesis focuses on a challenging problem on how to efficiently and effectively recommend friends for users on social networking sites. Due to the varieties of social networking services, we divide social networking services into social-based social networking services and information-based social networking services according to the usage purposes of different users. Users’ main purpose of using social-based social networking services is to communicate with offline friends or acquaintances, and users’ main purpose of using information-based social networking services is to look for interesting information. This thesis investigates a friend recommendation problem for these two types of social networking services respectively. In order to recommend friend efficiently and effectively on social networking sites, we firstly investigate the micro patterns of two types of social networking services as the theoretical basis for the designs of friend recommendation algorithms. Secondly, we design friend recommendation algorithms for two types of social networking services respectively. Lastly, we describe two applications based on our proposed friend recommendation algorithms. Specifically, we firstly explore whether the traditional theories on cost, reciprocal, triad closure and homophily in the physical world would be still valid in these two types of online social networking services. The exploration provides a theoretical basis for the subsequent designs of friend recommendation algorithms that should help enhance social networking services.Secondly, we propose three friend recommendation algorithms. For users of social-based social networking services, we propose a local random walk based friend recommendation approach by bringing together social network and tie strength. We firstly construct a weighted friend network as the basis for friend recommendation. Then, users’ similarity is determined by a local random walk based similarity measure on a weighted friend network. For users of information-based social networking services, this thesis proposes a memory-based friend recommendation approach and a model-based friend recommendation approach. The memory-based friend recommendation approach has the characteristics of real-time, intuitive and without offline modeling, and the model-based friend recommendation approach has the characteristics of higher accuracy. The memory-based approach combines collaborative filtering and link prediction methods. We firstly locate similar interest users of a target user by walking through certain topological structures. We then identify candidates whom the similar users are following. After identifying candidates of the target user, we rank candidates using unified weighting schemes. Finally we promote a group of selected candidates to the target user. The model-based approach essentially adapts the matrix factorization model in traditional item recommender systems to friend recommendation in information-based social networking services and uses structural regularization to exploit the structural information of a social network aimed at improving the accuracy of friend recommendation.Lastly, we describe two applications using our proposed friend recommendation algorithms. The first application is an e-commerce based social recommender by adopting the developed social-based friend recommendation algorithms. We propose an item recommendation approach which includes social-based friend recommendation algorithms to improve the accuracy of item recommendation in e-commerce. In the second application, we propose a social learning platform which incorporates information-based social networking services. We describe the system architecture of a social learning platform and propose algorithms to enhance the core services of the social learning platform by extending the information-based friend recommendation to recommend learning friends and resources for learners on the social learning platform, so as to improve the learning efficiency and enthusiasm of the learns on social learning platform.