RWR-Based Resources Recommendation on Weighted and Clustered Folksonomy Graph

Random Walk with Restarts has been proved as an effective model for collaborative recommendation in social systems, with ability to mitigate the problem of data sparsity. However, the present framework of RWR performs on un-weighted folksonomy graph, thus neglects some useful and implicit information inside the folksonomy, such as the preference of users to resources or tags, the awareness difference of users to resources of the same tag. Inspired by this, this paper presents a resources recommendation model which enhances the original RWR recommendation framework in the twofold. On one hand, the weights are assigned to the edges of folksonomy graph to indicate their importance. On the other hand, resource clustering is applied to solve the awareness differences of users. Experimental results on a Last fm dataset show that the new model can significantly improve the recommendation accuracy compared with original RWR-based recommending model.