WWW 2011 invited tutorial overview: latent variable models on the internet.

Graphical models are an effective tool for analyzing structured and relational data. In particular, they allow us to arrive at insights that are implicit, i.e. latent in the data. Dealing with such data on the internet poses a range of challenges. Firstly, the sheer size renders many well-known inference algorithms infeasible. Secondly, the problems arising on the internet do not always fit well into the known categories for latent variable inference such as Latent Dirichlet Allocation or clustering. In this tutorial we address a number of aspects. Firstly, we present a variety of applications ranging from general purpose document analysis, ideology detection, clustering of sequential data, and dynamic user profiling to recommender systems and data integration. Secondly we give an overview over a number of popular models such as mixture models, topic models, nonparametric variants of temporal dependence, and an integrated analysis and clustering approach, all of which can be used to solve a range of data analysis problems at hand. Thirdly, we present a range of sampling based algorithms for large scale distributed inference using multicore systems and clusters of workstations.