Strategies in Computer-Assisted Text Analysis

This paper reviews the logic of attempts to automate the processes involved in computer-assisted text analysis in the social sciences. Bayesian estimation methods in spatial analysis of variations in positions of political parties over time and Latent Dirichlet Allocation from the developing field of latent topic analysis are compared with the analysis of structures of word co-occurrences in the tradition of content analysis, using Procrustean individual differences scaling. Each depends in practice on concentrating attention on a limited number of word tokens regarded as meaningful while most are disregarded as inessential. By applying apparently competing strategies to the same set of party contributions to the 1997 budget debate in the Italian parliament, they can be shown to be complementary in character and should be applied as such in comparing material of this kind.