Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

TASER: Table Agents for Schema-guided Extraction and Recommendation

Created by
  • Haebom

Author

Nicole Cho, Kirsty Fielding, William Watson, Sumitra Ganesh, Manuela Veloso

Outline

This paper proposes TASER (Table Agents for Schema-guided Extraction and Recommendation), an agent-based system for extracting unstructured, multi-page table data from real-world financial documents. TASER transforms unstructured tables into regularized, schema-compliant output by utilizing agents that perform table detection, classification, extraction, and schema modification suggestions. Specifically, TASER incorporates schema improvements through continuous learning, emphasizes the effectiveness of large-scale batch learning, and achieves 10.1% performance improvement over existing models such as Table Transformer. Furthermore, we present a novel financial table dataset, TASERTab, which comprises 22,584 pages (28,150,449 tokens), 3,213 tables, and a total of $731,685,511,687 worth of asset data.

Takeaways, Limitations

•
Takeaways:
◦
Provides an effective solution to the problem of extracting complex and unstructured table data from real-world financial documents.
◦
Proving the Effectiveness of an Agent-Based, Schema-Guided Extraction System
◦
Emphasize the importance of performance improvement and schema improvement through continuous learning.
◦
Enabling research by releasing a large-scale dataset, TASERTab, including real-world financial data.
◦
10.1% performance improvement over Table Transformer
◦
Improved schema recommendations and increased asset extraction through large-scale batch learning (9.8%).
•
Limitations:
◦
Currently available information is insufficient to provide a detailed description of the specific architecture and algorithms of the TASER system.
◦
Further analysis of the quality and bias of the TASERTab dataset is needed.
◦
Generalization performance evaluation is required for various types of financial documents and table structures.
◦
Lack of comparative analysis with other agent-based systems.
👍
Made with Slashpage