This study aims to design, implement, and validate a user-centered platform integrated with artificial intelligence (AI) to mitigate information asymmetry and non-standardized transaction procedures in the used construction machinery market and to imp...
This study aims to design, implement, and validate a user-centered platform integrated with artificial intelligence (AI) to mitigate information asymmetry and non-standardized transaction procedures in the used construction machinery market and to improve the user experience (UX), particularly for middle-aged users. Although the domestic used construction machinery market is large and centered on high-value assets, it faces challenges in establishing transaction trust due to a broker-centered distribution structure and the accumulation of unstructured data.
To ensure both technical completeness and user acceptance, this study adopts the Design Science Research (DSR) methodology to develop the system architecture. The proposed platform was implemented with a FastAPI-based backend server and a Flutter-based cross-platform mobile application. In addition, a relational database model using MariaDB was designed to support stable data management and scalability.
To standardize listing data, which is a core requirement of the platform, an image-based automatic registration system was developed. Equipment images captured by users are processed by a YOLOv9 object detection model to extract equipment regions, and the extracted regions are then fed into a ResNet18-based multi-head classifier to automatically identify equipment category, manufacturer, and model. The identified attributes are mapped to structured database fields, transforming unstructured data into standardized transaction information and minimizing manual input burden.
For advanced AI functionality, the price prediction engine was evaluated using a Random Forest regression model and an MLP (Multi-Layer Perceptron) deep learning model. Experimental results confirmed that the Random Forest model achieved a coefficient of determination of 0.98, providing an objective basis for predicting actual transaction prices. The recommendation module combines ALS (Alternating Least Squares) collaborative filtering based on user behavior logs (e.g., views and favorites) with a KoSimCSE-based hybrid recommendation to address data sparsity and cold-start issues. In addition, a Bandit algorithm was applied to incorporate real-time rewards and optimize exposure strategies for personalized listings. An AutoEncoder-based unsupervised learning model was also applied to detect abnormal trading patterns that deviate from typical price ranges, thereby enhancing transaction safety.
The UI/UX was designed in accordance with ISO 9241-210 user-centered design (UCD) principles. Along with a Mobile-First strategy, an explainability principle (Explainable AI, XAI) was incorporated to visualize the rationale behind AI decisions. Usability testing indicated that the SUS (System Usability Scale) score improved after system refinement and that feedback on AI recognition results and a simplified interface enhanced accessibility for middle-aged users.
The expected contributions of this study are as follows: (1) improving industrial data management efficiency through automated pipelines, (2) minimizing development and maintenance costs by leveraging an open-source-based technology stack, and (3) providing practical guidelines that can reduce information asymmetry, increase transaction success rates, and accelerate digital transformation in high-value industrial asset markets.