Emails Classification: Comparing Statistics, Machine-Learning, Deep-Learning, and ChatGPT Prompting Techniques

The automatic classification of emails in a real scenario is very helpful for administration purposes and focuses human attention on solving tasks instead of organizing them. This paper presents a study of the applicability of a wide variety of approaches to the problem of this automatic email classification. For this purpose, we have conducted a wide set of experiments using traditional classification methods such as Support Vector Machines (SVM), distinct Large Language Models (LLMs), and prompting techniques applied to ChatGPT to tackle the classification problem. Thus, a set of labeled email conversations from LIS DATA Solutions have been studied, pre-processed, and prepared in order to be used by the different classification techniques. The results show that, surprisingly, traditional approaches beat more robust solutions for this task.