12.2 Topic models

  • Topic models…
    • …may provide classifications that are hard to interpret (especially topic models)
      • Remember that we have to label the categories/topics ourselves
  • Project: Measuring trust (Landesvatter & Bauer)
    • Aim: Classify short open-ended responses
    • “Simple” topic models (LDA, STM) did not yield satisfying results, hence, need to resort to supervised classification
    • Random forests (part of tree-based methods) turned out to be a viable way