Web Application for Large Language Model-based Diagnostic Analysis of Correlation Maps

The process of diagnostic analysis of data often incorporates human experts for problems outside of computer science and data analysis itself. These experts then interpret and explain the events, trends, and relationships identified in the studied data based on their previous domain knowledge. The main insufficiency of this approach to diagnostic data analysis is the lack of experts or their frequent unavailability, which motivated the previous use of pre-trained large language models as a surrogate for human experts in solving simple domain-specific problems. In this work, the web-focused application for diagnostic analysis of data based on the combination of large language models and correlation analysis is designed, implemented, and examined on two case studies of commonly utilized benchmarking datasets. The proposed model emphasizes the use of interactive visualization techniques in the context of correlation maps, in which the significant relationships between the value of attributes of a dataset are identified, while the large language model offers a brief explanation of these relationships from the point of view of the data domain.