Research Project

Table Copilot

A Table Assistant Empowered by Natural Language Conditional Table Discovery

SELECT * FROM tables WHERE condition = "natural language";
Table Copilot Interface
NLCTD

About Table Copilot

Table Copilot is an end-to-end table assistant featuring precise, interactive, and personalized table search capabilities. With the rise of large language models (LLMs), we've developed a solution that integrates table discovery directly into the table assistance process.

While current table assistants often overlook the initial step of table discovery, our aim is to create a seamless experience where users can search for tables using natural language conditions and perform complex operations with ease.

NLCTD Natural Language Conditional Table Discovery
LLM Powered by Advanced Language Models
E2E End-to-End Table Assistant

Key Features

Discover what makes Table Copilot powerful and intuitive

Natural Language Conditional Table Search

Search for tables using natural language queries with specific conditions (e.g., "CS grade above 90") without needing complex query syntax.

Advanced Table Fusion

Novel fusion-based solutions for handling both natural language conditions and query tables as input.

Intelligent LLM Integration

Carefully designed prompts enable LLMs to recognize table discovery tasks and invoke appropriate methods.

Multi-Interface System

Access table repository, table search, and table processing through intuitive interfaces.

Research Demonstration

See Table Copilot in action

This demonstration showcases Table Copilot's three main interfaces: the table repository, table search, and table processing. We illustrate how users can leverage natural language queries to discover relevant tables, apply conditions to refine results, and perform complex table manipulations.

Our Solution

How we address the challenge of table discovery

Table Copilot Architecture

We propose Table Copilot, an LLM-based table assistant capable of advanced table discovery. Our approach includes:

  • NLCTD Definition: We define a new table discovery scenario named Natural Language Conditional Table Discovery (NLCTD) that accepts both natural language conditions and query tables as input.
  • LLM Integration: We prompt LLMs with carefully designed instructions to recognize table discovery tasks and invoke appropriate discovery methods.
  • Fusion-Based Solution: Our novel fusion method makes table discovery more intuitive and user-friendly.
  • Three-Interface System: Table repository for data management, table search for discovery, and table processing for operations.

Contact Us

Get in touch with our research team

cuilingxi.cs@zju.edu.cn

django040805@zju.edu.cn

College of Computer Science and Technology, Zhejiang University