Learning Bayes Nets for Relational Data With Link Uncertainty Extended Abstract

We present an algorithm for learning correlations among link types and node attributes in relational data that represent complex heterogeneous networks. The link correlations are represented in a Bayes net structure. The current state of the art algorithm for learning relational Bayes nets captures only correlations among entity attributes given the existence of links among entities. The models described in this paper capture a wider class of correlations that involve uncertainty about the link structure. Our base line method learns a Bayes net from join tables directly. This is a statistically powerful procedure that finds many correlations, but does not scale well to larger datasets. We compare join table search with a hierarchical search strategy. A key challenge for relational learning that scales with data size is to compute event counts in a relational database (sufficient statistics), especially when these involve negated relationships. We describe how the fast Möbius transform provides a scalable solution for this problem.