The power of social media is often taken for granted. We know that it has been used to advertise, to politicize, to inform, and to educate. But at times such information seems to be targeted at a particular audience, so that one gets the precise restaurant, movie, or product recommendation that is needed. How does such targeting come about?

This paper posits that what they call online social network (OSN) sites are useful for a wide variety of activities, but that such information is arrived at in a particular manner that matches a source with the target through a system of trust. That trust is made possible by significant factors such as the degree of social intimacy, the impact of the community, the similarity of preferences, and the distance of residential location, among others.

The authors acknowledge previous work that has been done in the field, but make the claim that their proposed method is faster, more efficient, and more accurate. Beyond the social context within which individuals operate, there are independent social environmental factors such as social position, preferences, and residential location that are taken into consideration. Likewise, there are also dependent social factors such as social relationships and what they call the indegree (“the number of participants who have interactions with [this actor]”) and outdegree (“the number of other participants with whom this person has social interactions”) factors. Beyond those, there are contextual impact factors that come into play: trust, the degree of social intimacy, the impact of the community, the similarity of preferences, and the distance of the location of residence previously mentioned. This degree of relationship is arrived at through data mining techniques.

The authors’ heuristic social context-aware trust network extraction (H-SCAN-K) method goes beyond the transitivity of trust when they try to optimize their strategy, that is, “neglect some marginal nodes that also have high likelihood to connect to the target” and “select some marginal nodes that have low likelihood to connect to the target.” Thus, H-SCAN-K undertakes a bidirectional search--forward search and backward search--that enhances the quality of the trust network.

The authors use tables, graphs, figures, and formulas to demonstrate their point. They then apply their system to two real datasets of social networks: the Enron email dataset with more than 80,000 participants (nodes) and more than 30,000 links, and the Epinions dataset with more than 88,000 participants and more than 700,000 links, with the links being formed by trust relations specified by the participants themselves. Except for the paper’s typographical errors, such as references to the “1960’s” rather than “1960s” and problems with subject–verb agreement, the authors recognize that more needs to be done in terms of applying their models and algorithms beyond today’s social media, that is, to the next generation of social networks.