Classification of Dialogue Components in Viva Interactions Using BERT and SVM

Manual transcript analysis of dialog elements during Viva examinations turns into extensive labor which produces inconsistent results while generating unfair and scalable evaluation outcomes. The research study introduces an automatic machine learning system for viva dialogue classification to solve related limitations. A process of speech-to-text conversion occurs before natural language processing techniques extract features from the data. This work employs Support Vector Machines (SVM) for their effectiveness in high-dimensional space. BERT embeddings enhance contextual representation, improving classification accuracy to 0.89 for dialogue elements. The automated assessment system decreases workload while limiting human judgment processes to deliver consistent evaluations.