Exploring Text Similarity in Human and AI-Generated Scientific Abstracts: A Comprehensive Analysis

The rise of artificial intelligence (AI) and its ability to generate human-like text has sparked significant interest in understanding the similarities between human-written and AI-generated content. This project explores the text similarity between human and AI-generated scientific abstracts through a comprehensive analysis. The research focuses on comparing the quality and structure of abstracts written by humans and AI models using three deep learning-based algorithms: RoBERTa+BiLSTM, DistilBERT, and BERT+CNN. These models were trained to classify abstracts into two categories: Human Abstract and AI Abstract. By utilizing state-of-the-art natural language processing techniques, the models are evaluated on their ability to capture semantic and syntactic similarities between the two types of abstracts. The project highlights the effectiveness of hybrid models combining pre-trained transformer models with recurrent neural networks and convolutional networks for this classification task. Performance metrics such as accuracy, precision, recall, and F1-score are used to assess the models, with the goal of providing deeper insights into how AI-generated text compares to human writing in scientific contexts. The results of this analysis can offer valuable implications for the development of AI-driven tools in academic writing and contribute to the ongoing discussion on the authenticity and reliability of AI-generated content in scholarly research.

Keywords: AI-generated text, Text similarity, Scientific abstracts, RoBERTa+BiLSTM, DistilBERT, BERT+CNN, Natural Language Processing, Deep Learning, Text classification, Hybrid models.