Project Summary

The Social Web has originated an unprecedented growth in the use of social networks, folksonomies, blogs, wikis and similar systems. These tools encourage collaboration and exchange of information among users, creating a huge and growing amount of digital resources that often overwhelms the human processing capabilities. To address this problem, recommender systems filter the available information, and suggest the user those resources he may be interested in, without the need of explicitly looking for them.

The use of classical recommendation models has been successful in a wide range of e-commerce portals. These models, however, require further improvements. ReSHeT project addresses two of these improvements: the consideration of information heterogeneity and the inclusion of the time dimension.

Specifically, we propose to develop recommendation models that exploit user preferences from several sources (ratings, annotations, resource consumptions, social networks, etc.), and which can be applied to resources belonging to different media (text, image, audio, video). By testing these models on datasets obtained from real social systems, we shall analyse factors that are influenced by time (e.g. changes in user preferences, differences among the user’s preferences according to the time of day or the month of the year, resources subject to fashions and trends, etc.). Based on such analysis, we shall propose and evaluate a strategy to change or adjust the developed recommendation models over time.