Reveal Presents at JSM

Reveal is pleased to announce its participation in the Joint Statistical Meetings (JSM) 2024, held in Portland, Oregon. As one of the largest statistical conferences in the world, JSM brings together thought leaders, researchers, and industry professionals to discuss advancements in statistical science and data innovation.

At this year’s conference, Reveal presented its latest research paper, Ensemble Retrieval Strategies for an Improved NAICS Search Engine in the U.S. Census Bureau, co-authored by Cameron Milne, Yezzi Angi Lee, Taylor Wilson, and Hector Ferronato. This groundbreaking study explores the use of Large Language Models (LLMs) and semantic retrieval techniques to enhance the accuracy and efficiency of the North American Industry Classification System (NAICS) search engine.

The research investigates five key questions related to multi-stage retrieval pipelines, demonstrating how semantic search significantly outperforms traditional keyword-based approaches. By leveraging advanced retrieval strategies, the team achieved an accuracy of 86.87% on synthetic data—a 45.1% improvement over existing keyword search methods. This work offers valuable insights for optimizing information retrieval in federal data systems and other large-scale classification applications.

Our participation at JSM 2024 highlights Reveal’s commitment to advancing statistical methodologies and data-driven solutions. By applying cutting-edge AI technologies, we aim to improve the accessibility and usability of critical economic data, ensuring that statistical systems better serve policymakers, researchers, and the public.

Reveal remains dedicated to developing innovative solutions that enhance data accessibility and accuracy across industries. We look forward to continued collaboration with the U.S. Census Bureau and other federal agencies to push the boundaries of AI-driven information retrieval.

For more information on our research, please visit the conference proceedings or contact us directly.