ST-Raptor is a new agentic system designed to address the challenges of question answering (QA) on semi-structured tables. These types of tables, often used in business and scientific contexts, require precise extraction of cell contents and positions, as well as correct interpretation of implicit logical relationships.

Architecture and Functionality

ST-Raptor combines several techniques to improve accuracy and usability. It includes an interactive analysis environment with visual editing, tree-based structural modeling, and agent-driven query resolution. This approach aims to overcome the limitations of traditional methods, such as converting tables to structured formats (which can cause information loss) or using multimodal models that struggle with complex layouts.

Performance and Availability

Experimental results on benchmark and real-world datasets demonstrate that ST-Raptor outperforms existing methods in both accuracy and usability. The project code is available on GitHub, and a demonstration video is available on YouTube.