Measurement and analysis of online social networks
Proceedings of the 7th ACM SIGCOMM conference on Internet measurement
Crawling Facebook for social network analysis purposes
Proceedings of the International Conference on Web Intelligence, Mining and Semantics
Computing communities in large networks using random walks
ISCIS'05 Proceedings of the 20th international conference on Computer and Information Sciences
| Hi-index | 0.00 |
The advent of business oriented and social networking sites on the Internet have seen a huge increase in number of people using them in recent years. With the expansion of Web 2.0, new types of websites have emerged such as online social networks, blogs and wikis. Their popularity has resulted in exponential growth of information on the web and interactions overload thus making it harder to access useful or relevant information. Recommender systems are one of the applications employed to address this problem by filtering relevant information and enhancing user experience. They traditionally use either the content of items of the websites (content-filtering recommender systems) or the collaboration between the users and items such as rating (collaborative-filtering recommender systems) or a combination of them (hybrid recommender systems). However due to the nature of data they use, they all have one or more weaknesses such as cold start, sparsity of data, scalability problems and overspecialised recommendation. Social networks and other similar websites have new types of data which can be used in recommender systems thus have the potential to overcome these shortcomings. However without a good understanding of the properties and structure of these online social websites, the applications can not be accurate. This paper presents an empirical measurement study of the properties and structure of one such social websites. It examines an online movie database, and the interactions between reviewers and attempts to construct a social network graph based on the network of reviewers. The resulting network is confirmed as the power-law, small-world and scale-free. It identifies the highly connected clusters and shows that the content of these subgroups are diversified and not limited to similar tags. Finally the implication of these finding is discussed in order to enhance current recommender systems enabling them to provide diverse results while overcome their shortcomings.