Akademik Veri Yönetim Sistemi
7th International Conference on Innovative Academic Studies, Konya, Türkiye, 2 - 03 Eylül 2025, ss.72-77, (Tam Metin Bildiri)
As a key route to resource efficiency and the circular economy, industrial symbiosis (IS), which encourages the reuse of waste, byproducts, and energy across firms, has grown in significance. However, the complexity of sectoral relationships, a lack of collaboration, and a lack of data make it difficult to identify possible IS opportunities. There are now more options for automating this discovery process thanks to recent developments in large language models (LLMs), especially ChatGPT. With an emphasis on matching waste codes (EWC) to pertinent sector codes (NACE), this study examines how prompt engineering techniques affect how well LLMs identify IS opportunities. Three prompt strategies—zeroshot (original), role assignment, and chain-of-thought (CoT)—were compared using eleven EWC codes. The ChatGPT-4o paid version from July 2025 was used to test each prompt. According to the findings, role assignment and CoT only produced three and one matches, respectively, whereas zero-shot prompting produced the most valid IS suggestions (4 out of 11). This demonstrates the erratic nature of LLM reasoning in technical domains and suggests that more intricate prompting does not always result in better performance. According to the results, LLMs can help with early IS investigation, but they should only be applied sparingly and in tandem with professional validation. To improve practical reliability and policy relevance in circular economy planning, future research should investigate hybrid prompting strategies, model fine-tuning with IS-specific corpora, and integration with industrial databases.