Social networking is crucial for developing networks of common interest, but it is important to remember there may be costs involved. In this paper, the deployment of the Flickr application programming interface (API) is discussed. Wang et al. present a probabilistic model intended to uncover latent topics of interest among Flickr users and groups. The purpose is to improve the recommendation of groups to users and users to groups. Probability models are useful for developing inferences from large datasets, and are frequently used in decision support systems.

The authors list related work, and describe their own model as “a hybrid approach [that] exploits both visual contents and the existing links between users and Flickr groups.” The initial phase of this work seeks to discover common topics of interest among users and groups, the claim being that the results yield a more consistent dataset. To that end, the authors undertook an examination of visual and textual features using a vocabulary of visual words [1] in conjunction with the textual tags associated with each image in the dataset.

Visual data generates large amounts of overhead, and the analysis of visual data is complex. The comparison of the published results of this study with other methods of group recommendation discussed here suggests only an incremental improvement in topic/user/group matching. Although this work is interesting, I am not convinced the authors have made substantial progress in predicting latent or other Flickr topics. However, further work may yield results that are more promising.