ChatGpt ile Veri Üretimi: Duygu Analizi Sınıflandırması Üzerine Bir İnceleme Data Generation with ChatGpt: A Review on Sentiment Analysis Classification

Imbalanced datasets are encountered in many real-world scenarios and pose a significant challenge, especially in machine learning applications. This challenge arises from the numerical underrepresentation or overrepresentation of certain classes or categories, which can negatively affect the generalization capability and performance of predictive models. This study focuses on two separate and imbalanced datasets, each with three different sentiment labels: negative, neutral, and positive. To address the imbalance caused by numerical differences among labels, data for the underrepresented label categories were synthetically generated using the Synthetic Minority Over-sampling Technique (SMOTE) and Generative Pre-trained Transformer (GPT). Subsequently, various text classification models were applied to these expanded datasets, and the performance of each model was comparatively analyzed. The primary aim of the research is to analyze the impact of imbalanced datasets on classification algorithms and to evaluate the potential positive effects of data augmentation performed using GPT on the performance of these algorithms. The positive impact of GPT-generated data on model successes promises hope for future studies.