The lecture started with Jeff giving his related works presentation on predicting eventual friendships. Through the use of cultural algorithms, katz centrality, and social theory, he wants to predict who you will become friends with. He finds that one of the key problems is current methods only account for a 1 to 2 hop distance in the network. By this account, you could only become friends with a friend of a friend. This, of course, is just a brief overview of some of what Jeff covered in class. For more information, check out his slides on the main CS 790g website.
After that, Dr. Gunes gave a lecture on Graph Data Mining. As mentioned in class, graph pattern mining can be used to do a number of things, including but not limited to: mining biochemical structures, program control flow analysis, and mining XML structures or web communities to name a few. A few of the graph mining algorithms mentioned in the lecture are Incomplete Beam Search, Inductive Logic Programming, and Graph theory-based approaches (this included the Apriori-based approach and the pattern-growth approach.) We covered the two graph theory-based approaches in class.
Two important definitions to remember from this lecture were frequent and support. The support of a graph g is pretty much how often the graph occurs within the larger structure. Meanwhile, a graph (or subgraph) is considered frequent if its support in a given dataset is no less than a minimum support threshold.
No comments:
Post a Comment