An AI-Based Performance Evaluation System for Oral Presentations

Resumo

Scoring plays a crucial role in evaluating performance, particularly in academic and professional settings. However, the assessment of oral presentations remains largely subjective and time-consuming, with limited automated solutions available. While automated essay scoring systems have been extensively studied, comparatively little attention has been given to the evaluation of spoken presentations. This work addresses this gap by proposing an AI-based grading system designed specifically for oral presentation assessment. The system integrates three transformer-based models, Sentence-BERT for measuring contextual relevance, KeyBERT for keyword alignment, and RoBERTa for coherence analysis to provide a comprehensive evaluation of spoken content. Experimental results show that the proposed approach achieves an average score deviation of less than 5% when compared with internationally accepted human evaluation standards, demonstrating strong reliability and consistency. Overall, the system offers a scalable, objective, and efficient solution for assessing oral presentations across varying levels of performance.