Learning statistical models from relational data

This thesis describes our approach to learning statistical models from relational data. Our goal is to build statistical models that are able to capture the important inter-table correlations in the data and exploit the information available in relational structure of the data. These statistical models can more accurately capture the dependencies and correlations in the domain than previous approaches and prove useful for both data exploration and summarization in relational domains. Our contributions include a collection of statistical models applicable in relational settings and automated induction algorithms for the models. In addition to the theoretical descriptions of the models and learning algorithms, we demonstrate their application on several real-world databases.