Blockmodeling and Text Classification

In this paper I discuss the application of stochastic blockmodeling to the domain of legal opinions, and specifically to the end of classifying those legal opinions into topics. I begin by defining the basic social network problem and the stochastic blockmodel in Section 2. In Section 3 I introduce the problem of labeling legal opinions and discuss the particular features of this domain. Section 4 describes the supervised methods that I will use. Then Section 5 shows how blockmodeling can be used with the citation information from legal opinions and presents evidence that it captures their natural relational structure. In Section 6 I combine blockmodeling with supervised methods to improve the resulting classifiers and give experimental results. Finally, Section 7 compares this method to other ways of using citation structure.