I'm really interested in how the personalization stuff and link graph are merged at Google. Part of what we're doing at Directed Edge is merging graph-based (P

I'm really interested in how the personalization stuff and link graph are merged at Google. Part of what we're doing at Directed Edge is merging graph-based (PageRank and HITS-like algorithms) and matrix-based (collaborative filtering) into a single model that will allow us to do information filtering for recommendations. Once you've got both of them loaded up onto the same model you can do some interesting things clustering wise. One of the biggest problem with typical matrix clustering, i.e. what Google News uses, is that you're limited in trying to move out to neighboring clusters. If you combine both of them into a single graph model you've got more predictive power in terms of what you expect for a user to know about and what you expect to be new information for them.