Sentiment Mining in ChatGPT Dataset: Leveraging Rule-Based and Transformer-Based Models

The rapid popularity of ChatGPT and conversational AI systems has sparked considerable interest in comprehending opinions expressed on social media platforms. This research analyzes sentiment analysis in datasets associated with ChatGPT, utilizing both rulebased models and sophisticated transformer-based architectures. A comparative analysis employs VADER, a lexicon-based method, and advanced transformer models including BERT, DistilBERT, and RoBERTa, in conjunction with RNN architectures such as LSTM and Bi-LSTM. The findings indicate that transformer-based models surpass conventional and RNN-based approaches, with RoBERTa attaining the highest accuracy of $\mathbf{9 8 . 7 6 \%}$ owing to its capacity to comprehend contextual subtleties. VADER demonstrates proficiency in rapid, rule-based sentiment analysis but exhibits deficiencies in accuracy inside intricate contexts. Exploratory Data Analysis (EDA) yields valueable insights into sentiment distribution by categorizing sentiments as positive, neutral, or negative, and also generates word clouds for each category of tweets, thereby illustrating user involvement with ChatGPT. This research not only illustrates the effectiveness of advanced models in sentiment analysis but also establishes a framework for future investigations to enhance sentiment mining techniques for conversational AI datasets, with potential applications in opinion monitoring, user feedback analysis, and real-time trend identification.