ABSTRACT
Evaluating subjective, open-ended answers is very difficult to do in manually. Existing automated methods have limitations in truly understanding the actual meaning behind answers. To address this, this paper puts forward a new integrated approach using natural language processing and machine learning methodologies to evaluate score descriptive answers automatically. We assembled a dataset for this study that included more than a thousand subjective questions along with sample answers, student responses, and crucial keywords. The dataset was also annotated to facilitate training and assessments. The proposed approach Utilizes semantic techniques like word2vec to represent words in a way that Maintains meaning, and semantic measurement using word mover's distance similarity between answers, unlike conventional methods. Without using any machine learning model, the proposed techniques alone achieve remarkable 85% accuracy in automatically scoring answers, which further improves to 86.3% accuracy when adding a machine learning model. In this paper, we propose improved way which analyse written answers using advanced computer methods. The proposed techniques and annotated dataset have strong potential for real-world application in education and examinations that involve open-ended subjective questions.
