| Popis: |
Text mining methods are widely used to analyze a respondent’s open-endedresponses from questionnaires. The Latent Dirichlet Allocation (LDA) model is oneapproach that summarizes information from a set of open-ended questions by identifying underlying patterns of word co-occurrences, or topics. There have been several models that extend LDA by including external covariates in the model. We propose to extend LDA by jointly modeling the topics with latent traits used in item response models. We call the new model a constructed response model. The new approach allows researchers to gain a better understanding of the constructs one intends to measure. We describe a Bayesian approach to estimate both the item response and topic model and a set of simulation studies to examine the performance of the new model. An applied example is also provided. We conclude with recommendations and future directions. |
