Deep Grading: Incorporating Argument Strength into Neural Essay Evaluation

The automation of essay grading poses a significant challenge, especially when evaluating the aspects of argument strength in essays designed for discussion, debating and argumentation. Addressing this challenge, we developed a methodology for calculating argument strength, effectively reducing the impact of essay length variations. Leveraging this argument strength scoring methodology, we built a deep neural network model aimed at learning enhanced embedding representations for essays, capable of capturing the intricate nuances of argumentation, which encompass information beyond textual and structural characteristics. Our results demonstrate competitive performance, surpassing traditional and deep learning embedding baselines.