Application of Fuzzy Cognitive Map for Policy Simulation in the Agricultural Sector

Abstract

Agricultural socio-economic research, such as Rural Rapid Appraisal (RRA) or Participatory Rural Appraisal (PRA), often requires quick results while dealing with subjective data. RRA systematically collects rural information, while PRA emphasizes community participation and engagement. In this context, Fuzzy Cognitive Mapping (FCM) assists in managing subjective, ill-defined, and fuzzy variables. FCM creates a network of concepts within systems through positive and negative connections, integrating diverse stakeholder perspectives to support decision-making. This research presents a preliminary model using Large Language Models (LLM) to develop FCM for oil-palm farming, specifically examining the role of middlemen. The FCM is built on subjective variables gathered through open interviews with two palm-oil farmers in Central Kalimantan. Hebbian learning enhances the accuracy of relationships between variables. Results show that FCM effectively simulates policy impacts and their outcomes across different scenarios by adjusting the learning rate and the slope of squashing function. Future research should incorporate more stakeholders to enhance LLM's capability in mapping diverse perspectives.