This research systematically explores the Model Variability Problem (MVP) in LLM-based sentiment analysis, highlighting inconsistent classifications and uncertainty issues that impact reliability.

• Identifies key challenges: stochastic inference mechanisms, prompt sensitivity, and training data biases
• Analyzes how these factors create inconsistent sentiment classification and polarization
• Proposes mitigation strategies and emphasizes the role of explainability
• Presents a framework for more reliable sentiment analysis applications

In medical contexts, this research is crucial as healthcare requires highly reliable sentiment analysis for patient feedback, clinical documentation, and treatment decisions where inconsistency could impact care quality and safety.

An overview of model uncertainty and variability in LLM-based sentiment analysis. Challenges, mitigation strategies and the role of explainability