AUTOMED: A Finite Automata-Based Framework for Health Misinformation Detection in Medical Chatbots

The AUTOMED model proposes an innovative system for automated analysis of chatbots in the healthcare sector, particularly in cardiovascular disease. The framework, which has already demonstrated effectiveness in other natural language processing contexts, is a finite-state automata model that identifies suspicious linguistic patterns and potentially misleading content. Furthermore, for clinically correct patterns, the system provides a possible interpretation of the patient's symptoms, which could be helpful to the physician. This approach stands out for its transparency and interpretability compared to methods based solely on deep learning, offering a more reliable and verifiable solution in sensitive areas such as medical communication. The model involves creating an annotated corpus of medical texts generated by chatbots, labeled by experts, along with a linguistic analysis to detect ambiguous or misleading expressions. In this paper, we present the model and its implementation, based on a combination of string-matching and rule-based finite automata capable of recognizing rule-driven structures. To our knowledge, this is one of the first attempts to integrate formal language models into the analysis and validation workflows of medical chatbots.