GMR-124 Emotion-Based Synthetic Feature Binary Classification of Human vs LLM Generated Text/Essay

Sentimental analysis is a popular method to classify text into various emotional tones and intentions. In the meanwhile, the emergence of large language models (LLMs) has become ever more capable, their potential to cause harm through information fabrication, misleading propagation, or mere lack of capability has also increased. Therefore, our project is designed to discover any patterns that could potentially uncover texts origins of human and LLM during sentimental analysis. Our dataset covers over 57,000 lengthier essay samples (70% human vs 30% LLM), we use the state-of-art pre-trained DistilRoBERTa-base, a powerful pre-trained language model that is a more condensed and speedier version of Google's BERT, as our bidirectional transformer. Moreover, we plot the experiment results in histograms and statistical analysis and propose potential future research directions.