Suche

Following our earlier session on quantitative textual analysis, Sophie Mützel will introduce us to applications of topic models. Originally, developed in computer science, machine learning, and natural language processing, topic models have recently gained increased attention in digital humanities, but also in the social sciences. Based on a particular probabilistic distribution, topic model algorithms cluster together words into "topics" without a priori coding by analysts. Topic modeling thus can be used as a tool to discover themes in large collections of documents.

Our commonly used frequentist statistics tell us the likelihood to have randomly drawn our particular random dataset at hand, if in reality the null-hypothesis was true. This is quite complicated and so it comes at no surprise that we often falsely interpret these inferential statistics as the likelihood that the null-hypothesis is true, or even worse that a statistically significant finding is highly likely to be true. But is it not possible to infer the likelihood of our findings given the data rather than the likelihood of our data given the findings?