Improving NLP Tasks in Movie Review Analysis Using ChatGPT Data Generation-Based Augmentation

Natural language processing remains a prominent area of interest within deep learning., encompassing fields such as film review analysis where dataset availability poses significant challenges. The advent of Chat Generative Pre-Trained Transformer (ChatGPT) has expanded opportunities for leveraging datasets in novel ways. This study aims to propose utilizing ChatGPT to generate text and augment original datasets to enhance the training models. Two experiments were designed: the first experiment controlled the ratio of ChatGPT-generated data to raw data in the dataset used for model training., measuring metrics such as accuracy., precision., recall., AUROC., and epochs for optimal results. The second experiment focused on manipulating dataset size using only raw data., with identical metrics recorded as in the first experiment. Comparing the experimental outcomes revealed that the ChatGPT-generated data improves results particularly when original data is limited. Augmenting original data with generated datasets accelerates model convergence., thereby achieving faster model training objectives.