The social engagement enabled by the Internet moved critical evaluation beyond the specialized few and distributed it in the hands of the common many. It is routine to evaluate the reputation of items in an online shopping context by the five-star ratings, comments, photos, and responses of users, manufacturers, and storeowners. Collectively, we evaluate without considering the reputation of the individuals reviewing. While retailers have attempted to offer information to add weight to the potential trustworthiness of the reviewer (for example, top ten reviewer, most helpful review, total reviews, member since, and so on), it is the increase in information volume that is enabling shoppers to become more courageous. Might it be possible to improve upon this dynamic across all relevant online social contexts by automating the evaluation of user reputation and provenance to create even more trusting and compelling context? Ceolin et al. further this topic forward in this paper.

Ceolin et al. tackle the challenging domain of social tagging (folksonomy generation) around the collection of “museum-quality” web artifacts. Institutions have more content than they have specialized library science types to annotate, classify, and organize. Enabling the “crowds” to assist in this work potentially benefits the greater good, if the work is of sufficient quality and the effort to evaluate that is radically reduced and automated. Ceolin et al. share their method and algorithms with clarity, enabling others to reference them for inspiration or replication.

Among the interesting aspects of the authors’ work is their ability to not simply apply their combinatorial analytics of user reputation and provenance to the individual but to the persona clusters--or as they call them stereotypes--for even greater applicability of their approach. Given an individual’s behavior, his or her present behavior cluster may be enough to draw a conclusion, despite the lack of historical performance information. This is highly valuable in diverse user bases and in situations with less consistent historical data to work with.

Finally, the authors did a superb job considering the robustness of their design. Overall, their approach survives quite well with the sleeper and on-off attack representing the greatest risk of fallibility. As with any of these exploitations, context must be considered. In some environments, it would be worth no additional effort to defend against data poisoning; however, given such a reasonable evaluation, that assessment is left to the reader.

With such a large amount of socially annotated information online, more can be done to create greater levels of truth. Ceolin et al.’s work furthers this topic, presenting their approach in a novel, accessible way, wrestling with relevant topics applicable across domains.