Relational Markov Networks
Ben Taskar, Pieter Abbeel, Ming-fai Wong, Daphne Koller · The MIT Press eBooks · 2007
One of the key challenges for statistical relational learning is the design of a representation language that allows flexible modeling of complex relational interactions. Many of the formalisms presented in this book are based on the directed graphical models (probabilistic relational models, probabilistic entity-relationship models, Bayesian logic programs). In this chapter, we present a probabilistic modeling framework that builds on undirected graphical models (also known as Markov random fields or Markov networks). Undirected models address two limitations of the previous approach. First, undirected models do not impose the acyclicity constraint that hinders representation of many important relational dependencies in directed models. Second, undirected models are well suited for discriminative training, where we optimize the conditional likelihood of the labels given the features, which generally improves classification accuracy. We show how to train these models effectively, and how to use approximate probabilistic inference over the learned model for collective classification and link prediction. We provide experimental results on hypertext and social network domains, showing that accuracy can be significantly