AI-driven mock interview assessment: leveraging generative language models for automated evaluation

In the education sector, adaptive support is critical for every student to face open-ended activities that need behavioral change, performance, and a pro-active learning mindset. This can be accomplished by using brilliant learning environments powered by artificial intelligence. Timely feedback is critical for helping students enhance their overall personality in learning, confidence, communication, and problem-solving. It is a highly demanding task for every teacher to conduct a mock interview and then provide feedback for each criterion. It is time consuming and may be delayed. However automated review of mock interviews can give timely student feedback while reducing the manual evaluation burden on teachers in areas with a high teacher-to-student ratio. Current ways of analyzing student interview responses include transformer-based natural language processing models, which have various degrees of effectiveness. One major problem in training these models the need of more data, as most of the datasets are based on HR queries, which have sufficient datasets. But, none of the interview recordings included both HR and TR-related questions. We recorded mock interviews with undergraduate students from multiple backgrounds to address the data scarcity issue. We extracted audio from the recordings, followed by transcripts that included speaker identification. We split the speakers' data into questions and responses. The biggest challenge is evaluating these answers, given the need for appropriate datasets for technical questions. This article investigates the text-generating AI model, GPT-3.5, to establish whether prompt-based text-generation approaches are viable for generating scores for specific student responses. Finally model results are compared with human values. Our results reveal that the pre-trained model yields excellent outcomes in interview grading.