Automated Framework for Civil Complaint Analysis in Road Construction Projects Using Natural Language Processing

Civil complaints are a significant risk in road construction projects owing to their inherent unpredictability and broad scope. Therefore, complaints must be managed systematically to facilitate proactive prevention and prompt responses. Currently, civil complaint data from national road construction projects in Korea are collected through civil complaint processing records and official documents, which are digitally archived via the Ministry of Land, Infrastructure, and Transport’s construction project management system. However, the absence of structured classification criteria complicates the retrieval of relevant information and hinders comprehensive analyses. Previous studies on construction civil complaint management have focused on analyzing complaints and providing information for citizens or administrative agencies, and a comprehensive approach to civil complaint management from the perspective of construction sites has not been established. Therefore, this study designed an automated framework that uses natural language processing (NLP) to analyze civil complaint documents related to road construction projects. Three task models were developed to extract and classify practical civil complaint information and an automated document analysis process was established. The models included KoELECTRA-based civil-complaint-type classification, civil-complaint-demand classification, and KPF-BERT-based facility-name recognition models. Data collection and preprocessing standards, civil complaint classification criteria, and a dictionary of facility terminology were developed to support the creation of the models. The weighted average F1 scores of the three task models were 0.936, 0.976, and 0.757, respectively. The proposed framework was implemented as a prototype web-based interface. This allowed key civil complaint information to be extracted rapidly and presented to support swift decision-making and systematically managed previously unstructured civil complaint documents. This study contributes to construction knowledge from theoretical and practical perspectives. It demonstrates the potential of automating road construction civil complaint analysis through an NLP-based structured framework. This framework also enables systematic and comprehensive management of complaint data, supporting complaint response operations at road construction sites.