Sign in Subscribe
Teachers

Recursive Method Unifies Information Extraction and Text Classification in Natural Language Understanding

Strategic Insights

— 1 min read

Researchers at Zhejiang University in Hangzhou, People's Republic of China, have proposed a new approach to unify information extraction (IE) and text classification (CLS) tasks in natural language understanding (NLU). By developing a recursive method with explicit schema instructor, the team aims to overcome existing limitations in NLU models. The study, published in the IEEE Transactions on Knowledge and Data Engineering, demonstrates the effectiveness and superiority of the proposed method, dubbed RexUniNLU, in various experiments conducted on English and Chinese datasets.

Key Takeaways:

  • The researchers redefined the concept of universal information extraction (UIE) with a formal formulation that encompasses almost all extraction schemas, including quadruples and quintuples.
  • They introduced RexUniNLU, an universal NLU solution that employs explicit schema constraints for IE and CLS, which encompasses all IE and CLS tasks and prevents incorrect connections between schema and input sequence.
  • The proposed method resets the position IDs and attention mask matrices to avoid interference between different schemas.
  • The team conducted extensive experiments on IE, CLS, and multi-modality, revealing the effectiveness and superiority of RexUniNLU.
  • The research has been peer-reviewed and published in IEEE Transactions on Knowledge and Data Engineering.

Statistics:

  • 37:11 issue of IEEE Transactions on Knowledge and Data Engineering (2025)
  • 6624-6635 pages in IEEE Transactions on Knowledge and Data Engineering (2025)
  • 2025: the year the research was conducted
  • 11: the issue number of the IEEE Transactions on Knowledge and Data Engineering where the research was published
  • 37: the volume number of the IEEE Transactions on Knowledge and Data Engineering where the research was published

Sources:

  • IEEE Transactions on Knowledge and Data Engineering, Volume 37, Issue 11 (2025): "Rexuninlu: Recursive Method With Explicit Schema Instructor for Universal Natural Language Understanding"
  • IEEE Computer Society, 10662 Los Vaqueros Circle, PO Box 3014, Los Alamitos, CA 90720-1314, USA
  • Yangyang Kang, Zhejiang University, College of Computer Science and Technology, Hangzhou 310027, People's Republic of China

Read more