Predicting Expert Judgment on the Quality of Automatically Generated Items

• In Event: Automatic Item Generation

Abstract

The goal of this study is to predict expert judgment of the quality of automatically generated items. We employed two approaches: a feature-engineering approach using random forest and gradient boosting classifiers and a representation learning approach using BERT. The best performing models were achieved with the feature engineering approach.

Authors

• Guher Gorgun, University of Alberta

• Ikkyu Choi

• Jiyun Zu, EDUCATIONAL TESTING SERVICE