Workshop on Learning Statistical Models from Relational Data, IJCAI-2003

ide powerful modeling component but are often limited to a "flat" file propositional domain representation where potential features are fixed-size attribute vectors. Often the manual process of preparing such attributes is costly and not obvious when more complex regularities are involved. We are developing a methodology which combines the strengths of classical statistical models with the higher expressivity of features automatically generated from a relational database. Our interest in statistical relational learning developed while working on modeling in CiteSeer , an online digital library of computer science papers. CiteSeer contains a rich set of relational data, including citation information, the text of titles, abstracts and documents, author names and affiliations, conference or journal names. Applications we have addressed include: i) prediction in social networks, e.g. link prediction: given two papers estimate whether they cite each other [Popescul and Ungar, 2003] ,