Title: NLP-Powered Industry Classification Based on Business Information
Abstract:
Industry classification is essential business information in commercial insurance for risk assessment, underwriting, pricing, policy customization, and marketing strategies. However, assigning businesses to explicit classifications is traditionally a labor-intensive and time-consuming task. Automating this process can significantly reduce manual effort and cost. In this study, we present an NLP-powered industry classification framework leveraging business textual descriptions and the North American Industry Classification System. We employ state-of-the-art hierarchical text classification models with graph neural networks to incorporate label semantics, hierarchy, and co-occurrence patterns. Additionally, we integrate semi-supervised and active learning strategies to utilize a large corpus of weakly labeled and unlabeled business data. The proposed approach also applies to other domains beyond commercial insurance, including marketing, investment, and public-sector planning (e.g., the U.S. Census Bureau and IRS).
Short Bio:
Changyue Hu is a Ph.D. candidate majoring in Actuarial Science at the University of Illinois at Urbana-Champaign. Before joining the doctoral program, she obtained her Master’s degree in Actuarial Science and Bachelor’s degree in Mathematics. Her research interests lie in machine learning applications in actuarial science. She has been a graduate supervisor of several InsurTech-related research projects at IRisk Lab. She has also received the SOA James C. Hickman Scholar Doctoral Stipend, the State Farm Actuarial Science Scholarship, the Esther Portnoy Actuarial Science Scholarship, and the Student Presentation Awards at the Actuarial Research Conferences.