Civil complaints submitted to the Seoul Metropolitan Government’s Eungdapso system have continued to increase in volume and complexity, including lengthy and unstructured texts, scanned petition documents, repetitive complaints, and malicious expressions. These trends have intensified administrative workloads and exposed limitations in complaint handling processes that rely heavily on manual review and individual experience.
This study proposes and empirically evaluates a generative AI–based approach to enhancing civil complaint handling using actual complaint data from the Eungdapso system. Generative AI technologies were applied across key processing stages—intake, summarization, classification, and response—to support malicious expression detection and refinement, summarization of complex complaints, OCR-based handling of scanned complaints, RAG-based department recommendation, and draft response generation for repetitive cases.
The results show that malicious expression handling, complaint summarization, and department recommendation achieved stable performance and are suitable for early-stage adoption as decision-support tools, while response generation requires stricter safeguards due to administrative risks. Based on these findings, the study proposes a phased adoption strategy with operational guardrails, providing a practical roadmap for responsible AI-assisted civil complaint handling.