Diagnostic Analysis Approach to Correlation Maps Through Large Language Models

Diagnostic data analysis is a methodological process aimed at the identification of underlying causes of events, the effect of which is the presence of trends and patterns in a multidimensional dataset. Since in most cases this cause originates in the area described by the data itself, knowledge from the field of data analysis, computer science or statistics is rarely enough for the analyst to explain the patterns, and it is necessary to examine the field of data expertise itself. In this work, we focus on creating a model for diagnostic data analysis, which automatically describes naturally correlated pairs of attributes using a large language model and provides an output in the form of a Diagnostic Sheet Visualization. The proposed concept is then experimentally verified in case studies on two multidimensional datasets focused on environmental variable sensor measurements and cubic graph properties.