This dissertation offers two major contributions: (1) to evaluate the suitability of recommender algorithms for social networks. Such recommender algorithms may receive as input not only the social graph of these networks but also content-based data from recommended items.
For such, the relevant characteristics of social networks and the most important recommender techniques for these tasks will be surveyed. Special attention is given to the web-based system for social photo-sharing called Flickr and to the employment of visual metrics for image similarity.
The second contribution (2) is the construction of a framework for the modeling and analysis of social networks, as well as aiding the empirical study of recommender algorithms on these contexts. Also part of this framework are the best practices adopted throughout the work done on this dissertation, such as: techniques for the gathering, analysis and visualization of data; social networks classification; identification and modeling of recommending tasks within these contexts; implementation of algorithms and their architecture.
The relevance of such contributions lies on the enormous amount of information available online and on the ever-growing complexity of the relationships between this data. In this context, recommender systems may provide a great aid for end-users.
In this paper, we present a framework for specifying recommenders within the context of Social Media sites such as Flickr or Last.fm. Based on the standard SIOC ontology, we show how the various recommendation problems can be defined. We also present a general software framework for implementing recommenders based on this model framework, and show some results obtained by one recommender built using it.