Dependency Networks for Collaborative Filtering and Data Visualization

We describe a graphical representation of probabilistic relationships-an alternative to the Bayesian network-called a dependency network. Like a Bayesian network, a dependency network has a graph and a probability component. The graph component is a (cyclic) directed graph such that a node's parents render that node independent of all other nodes in the network. The probability component consists of the probability of a node given its parents for each node (as in a Bayesian network). We identify several basic properties of this representation, and describe its use in collaborative filtering (the task of predicting preferences) and the visualization of predictive relationships.