Developing a GPT-Based text Extraction Model for Cancer Information

By employing Aristotle's rhetoric as the theoretical framework, the present study aims to develop a model that automatically extracts the three key components of persuasive strategies-ethos (authority), pathos (emotional appeal), and logos (logic)-from answers to pertinent cancer questions on Quora, a social question and answer platform. Furthermore, we apply the model to discrete groups of the most upvoted and random (non-upvoted) answers to compare differences in the three persuasive components. The dataset consists of a total of 103 questions and their corresponding answers, including both upvoted and random answers. It was employed for preliminary findings, comprising a total of 33 questions and answers, with answers to 19 questions used as training data and answers to 14 questions used as test data. We annotated sentences in the answers according to the three types of rhetoric employed. We then fine-tuned models based on Generative Pretrained Transformers (GPT) to classify the phrases, achieving an average F1 score of 0.84. Paired sample t-tests confirmed our research hypotheses regarding ethos and logos, while our hypothesis about pathos was not confirmed. Results suggest that ethos and logos are effective in communicating cancer information to consumers, but that pathos is not.