AB
AB
AiBoss

All Categories

project

ST-Raptor - An AI-powered table-based question-answering tool that supports various semi-structured tables.

ST-Raptor is a tool for semi-structured table-based question answering. It requires only an Excel spreadsheet and a natural language question as input to produce accurate answers.

What is ST-Raptor?

ST-Raptor is a tool for semi-structured table question answering. It requires only an Excel-formatted table and a natural language question as input to produce accurate answers. The tool can handle various semi-structured table layouts and, combined with visual language models and tree-building algorithms, can be flexibly integrated with different large language models. ST-Raptor has a two-stage validation mechanism to ensure reliable results. ST-Raptor provides the SSTQA benchmark test, containing 102 tables and 764 questions, to evaluate its performance.

Main functions of ST-Raptor

  • Precise Questions and AnswersSimply input an Excel spreadsheet and a natural language question, and it can generate accurate answers.
  • Diverse form supportSupports various semi-structured table layouts, such as personal information tables, academic tables, and financial tables.
  • Multi-format inputSupports inputting tables from various formats such as Excel, HTML, Markdown, and CSV.
  • No fine-tuning requiredIt can be used without any additional fine-tuning of the model.

ST-Raptor's technical principles

  • Visual Language Model (VLM)Combined with a visual language model, it can understand and process visual information in tables.
  • Tree construction algorithm (HO-Tree)The algorithm analyzes and understands table structures through tree construction, thereby improving the ability to process complex tables.
  • Flexible integration of large language models (LLM)It supports flexible integration of different large-scale language models, such as Deepseek-V3 and GPT-4o, to improve question-answering performance.
  • Two-phase verification mechanismA two-stage verification mechanism ensures that the generated answers are accurate and reliable, avoiding the generation of incorrect answers.

ST-Raptor project address

  • GitHub repositoryhttps://github.com/weAIDB/ST-Raptor

Application scenarios of ST-Raptor

  • Corporate Financial ManagementFinance personnel can quickly obtain answers by entering budget forms, which helps control financial costs.
  • Academic research data managementResearchers can input experimental data into a table and query specific results, thus accelerating the research process.
  • Human Resource ManagementHR can input performance data into the performance report to check employee performance and provide support for management decisions.
  • Financial risk assessmentAnalysts input risk data into a table to identify high-risk customers and reduce credit risk.
  • Logistics and supply chain managementManagers can input logistics order forms to check inventory and transportation status, thus optimizing the supply chain.
Back