Project Discussion Section (April 30)
I inadvertently monopolized class today when asking if my project proposal was received. Sorry about that…
But we did have an interesting discussion as a result, or at least I think so. One specific thing we talked about was the question of how to inform probabilities used in link prediction using social network structure. We also discussed the problem in link prediction of deciding which of the nodes will receive a link once the prediction algorithm as decided that a new link will be formed. I particularlyly liked the dataset that was talked about (an online social network with attributes on nodes, and timestamps indication when who linked to whom. That sounds like a very useful set of data.
I was not aware that the Facebook API exposes attributes on nodes before (like the religion, race, education, etc. Professor Wu also proposed using keyword similarity metrics as another measure. He also mentioned that he has a student, Sean (last name unknown) who is working on the link prediction problem using this data. This definately seems relevant.
Next, we discussed another group’s topic proposal dealing with the friend of a friend project. What would be cool about this if it were an implementation project, would be to come up with a unifying ontology (or RDF vocabulary) that defined the common structures across online social networks like Myspace, Facebook, and LinkedIn, such that one could network-shift their social neighborhood across the different groups, or maybe query across all three (or more) platforms. Essentially, exporting the stuff the API lets you access into an RDF graph that is portable might allow a bridging application to exist outside the confines of the various platforms. That would be cool!
Finally, we ended the discussion with a sensor routing structure proposal, using PKI certificates and signing, such that you could credit people for serving as inter-community pathways. One question I have about that though, is that it seems hard to evaluate the resultant graph in the end. Do you get more incentives for being a router of more messages? And how does that work with the spamming-penalties we’ve discussed in Davis Social Links before?