The Effectiveness of a Dynamic Loss Function in Neural Network Based Automated Essay Scoring

Automated Essay Scoring (AES) is the task of assigning a score to free-form text (throughout this paper essay will be defined loosely to include short answers) using a computational system. The goal of AES is to mimic human scoring as closely as possible. The development of the Transformer in [1] has significantly improved the performance of Natural Language Processing (NLP) models to a point where it is achievable to use a purely neural approach to AES [2], [3]. This has created the possibility for many task-agnostic architectures and pre-training approaches which then allows for greater flexibility in the implementation of these models. This also makes the cutting-edge performance of these NLP models available for simple implementation in real world situations.