The article presents the Text2Analysis benchmark, a tool developed to fill the gap in advanced analysis like forecasting and chart generation in tabular data analysis. The existing research has primarily focused on basic tasks, hence neglecting these advanced analysis areas. Text2Analysis incorporates these advanced tasks and also includes unclear queries similar to real-world user questions. Five different annotation methods are developed to enhance data quality and quantity. The benchmark was tested on 2249 query-result pairs with 347 tables, and the results showed that it poses a significant challenge in the field of tabular data analysis, opening avenues for more advanced research.
Publication date: 22 Dec 2023
Project Page: N/A
Paper: https://arxiv.org/pdf/2312.13671
