Improved Semantic Features for Automated Essay Scoring with Hybrid Topic Modeling Approach

This thesis proposes a hybrid Automated Essay Scoring (AES) model with two new semantic features, content detection and sentence similarity towards prompt. For content detection, the work implements Latent Dirichlet Allocation (LDA) topic modelling algorithm to identify how well the content of a particular essay is in relation to all the other essays. For sentence similarity towards the prompt, the work implements the transformer model's embedding to determine if the essay is written surrounding a given prompt. The proposed model outperforms the existing state-of-the-art solutions in the long essay segment.