If one wants to recommend resources instead, the same principle used for tags can be applied. Note that if we use only the ?URY projection, we would end up at the standard user-based CF algorithm (see Eq. 19.3). But since tags can provide additional information about user interests, they can eventually boost the recommendation quality and thereby should be exploited. A trivial tag-aware recommender method is to compute the user neighborhood based on the ?UTY projection matrix and aggregate the resources of the neighborhood to generate the recommendation list. A similar idea is presented in [6], where first the user-tag projection matrix ?UTY is used to compute a ranked list of tags, whereby the recommendation list of resources is extracted. But by using only ?UTY alone, one discards the resource information, which in this case, is the key mode of interest. In this sense, one needs to find a way to accommodate all the three modes of the folksonomy in a 2-way data structure so that standard CF can be applied. Tso-Sutter et al. [37] proposed an approach for doing that by extending the typical user-resource matrix with tags as pseudo users and pseudo resources (see Figure 19.7). Note that in this way, the user/resource profile is automatically enriched with tags. A fusion algorithm is then proposed for combining user-based CF (ucf ) and item-based CF (icf ) predictions over the extended matrix. Recall that in the standard user-based CF for the resource prediction problem, the interestingness score of a given user u for a particular resource r is computed as the averaged number of neighbors that co-occur with resource r, i.e.,