Applying soft links to diversify video recommendations

In this paper we present a study of exploratory video search tasks and recommendation techniques based on a graph representation of past user com-munity interactions with the system, which have been used in a number of multimedia retrieval systems. We propose an extension for such graph-based usage representation techniques based on the creation of additional soft links between nodes. It is demonstrated how soft links can be incorporated into a graph-based representation and how different state of the art techniques can be adapted to use soft links. Our evaluation, based on a simulation-oriented technique and real interaction data gathered from users, shows how our soft links can help in improving the diversity and, in some cases, the accuracy of the studied recommendation techniques.