Official UD Coronavirus (COVID-19) Updates
  • Visit
  • Apply
  • Give
University of Delaware - Alfred Lerner College of Business & Economics

By Sunny Rosen November 20, 2019

University of Delaware professor of MIS and JPMorgan Chase fellow Xiao Fang, assistant professor of accounting Jing He and doctorate in financial services analytics (FSAN) student Xiaohang Zhao have won the best paper award at the 2019 INFORMS Workshop on Data Science, the premier academic conference for data science and business analytics.

Their paper was selected out of 62 submissions from a variety of prestigious institutes. These include universities in the United States, Switzerland, China, Israel and Singapore. After a rigorous peer review process, the UD team’s research emerged victorious.

“This award reflects the academic quality of the FSAN Ph.D. program.,” Fang said. “Our top students can compete favorably with those graduating from the nation’s top business Ph.D. programs. This paper also reflects the mission and theme of the FSAN Ph.D. program: developing data science methods to solve critical problems in the sector of financial services. “

In their winning paper, “A Deep Learning Approach to Industry Classification,” the team designed a novel text based industry classification system that used financial firms’ most closely related peers to classify them. The team assigned to each firm its most similar peers based on similarity scores, reflecting the proximity of the firms’ lines of business.

“In this sense, we designed a ‘peer-based industry classification system,’” Zhao said. “However, in order to calculate the similarity scores, we needed to assign a vector to each financial report, which is on average a long document featuring diversified topics. We designed a novel document embedding model that produced significantly better document vectors than existing models.”

The team’s model was capable of handling even long documents “with heterogeneous and drifting concepts,” the paper explained. The team was able to empirically demonstrate the superiority of this system, which they plan to comprehensively evaluate in the near future.