Learning Automated Essay Scoring Models Using Item-Response-Theory-Based Scores to Decrease Effects of Rater Biases

In automated essay scoring (AES), scores are automatically assigned to essays as an alternative grading by humans. Traditional AES typically relies on handcrafted features, whereas recent studies have proposed models based deep neural networks obviate the need for feature engineering. Those generally require training a large dataset of graded essays. However, grades in such known be biased owing effects rater characteristics when is conducted assigning few raters set each essay. Performance drops data used model training. Researchers fields educational and psychological measurement recently item response theory (IRT) that can estimate while considering biases. This study, therefore, proposes new method trains using IRT-based dealing with bias within data.