Approach to Unsupervised Instance Matching for Power Plants

This technical report presents a new algorithm named AutoCal, designed for unsupervised instance matching, particularly applied to linked data of power plants. The report outlines the significance of instance matching in populating knowledge graphs with high-quality data from diverse sources. AutoCal operates without the need for labeled data or method-specific parameter tuning, making it suitable for various domains. The algorithm's performance is compared with existing unsupervised matchers from machine learning and deep learning, demonstrating competitive results in matching quality while significantly improving runtime efficiency. The report details the evaluation of AutoCal through multiple test scenarios, emphasizing its capability to function effectively in automated environments. Additionally, it discusses the challenges faced in instance matching, including the necessity for accurate data preprocessing and schema matching. The findings suggest that AutoCal is well-suited for integration into systems like the World Avatar, which relies on accurate data for complex cross-domain tasks.