Cognition-enhanced geospatial decision framework integrating fuzzy FCA, surprisingly popular method, and a large...

Area of Science:

• Geographic Information Science
• Cognitive Science
• Artificial Intelligence

Background:

• Traditional geospatial analysis often overlooks cognitive influences on decision-making.
• Integrating human cognition can improve the accuracy and interpretability of spatial models.
• Large Language Models offer new possibilities for capturing cognitive biases in data.

Purpose of the Study:

• To introduce a cognition-enhanced framework for geospatial decision-making.
• To integrate Fuzzy Formal Concept Analysis (FCA) and a Large Language Model (GPT-4o) for capturing cognitive influences.
• To improve prediction accuracy and interpretability in geospatial analyses.

Main Methods:

• Fuzzy FCA was employed to extract concept hierarchies from spatial data.
• GPT-4o was utilized to estimate Surprisingly Popular (SP) scores, reflecting cognitive biases.
• Cognitively informed features were integrated into machine learning models (e.g., XGBoost).
• A novel method was developed to evaluate the cognitive validity of LLM-generated explanations.

Main Results:

• The cognition-enhanced framework significantly improved prediction accuracy and interpretability.
• XGBoost models achieved an accuracy of 0.8412 in real-world scenarios.
• The evaluation method confirmed the cognitive validity of LLM-generated explanations.
• The framework demonstrated effectiveness in urban mobility and environmental risk assessments.

Conclusions:

• Incorporating cognitive elements enhances geospatial model performance and interpretability.
• The framework bridges the gap between data-driven predictions and human decision-making.
• This approach has broad potential for GIS, urban planning, and environmental management.