Combining Language Models, deep ensemble learning and explanation for automatic ticket classification

In this work, we addressed the Ticket Classification task on the dataset provided by CINECA. Our goal was to provide a tool to reduce the workload on the customer care units, particularly by automating frequent and easily categorized tickets and offering easy-to-interpret analysis tools for more complex requests. After analyzing different approaches proposed in the literature, we developed our framework that managed to bring together interesting and original aspects namely the use of LM-based text representations, the implementation of ensemble models with different strategies of combining base learners, the analysis of the tradeoff between accuracy and confidence of the predictions and model complexity, and, finally, a mechanism of explainability of the results to mitigate the black-box nature of DNN models. Tests on real data confirmed the accuracy of the classifications returned by the framework and the practical value of their associated explanation.