IEEE ICPRS 2023

Transformer-based text classification for business analytics and public safety

Deep learning (DL) has received increasing attention in the last decade, becoming the de facto solution for several machine learning tasks, including computer vision, signal processing, or text analytics. For the latter application, attention mechanisms based on transformers have significantly improved the performance of DL models on natural language tasks. This gain has opened new possibilities for research in text analytics, creating new opportunities for improving decision-making. In this talk, we first focus on business analytics, automating the analysis of complaints, inquiries, and customer feedback. Providing a prompt answer to customer claims and complaints, as well as analyzing the factors that affect the customer experience the most, are current trends in modern marketing. For the second part of the talk, we present a NLP application for crime analytics. The main goal is to design robust DL classifiers in the presence of noisy labels. We consider data from a crime app, in which users categorize safety reports by choosing one of the various options provided by the app. The users, however, tend to provide incorrect labels either because it is easier for them, e.g. by choosing the first alternative, or by intentionally exaggerating the importance of their reports.

Carla Vairetti received her B.S. degree in Computer Science in 2000 from the University Nacional de La Plata, Argentina. She also received an M.S. degree in Sciences in 2013 from the Pontificia Universidad Católica, Chile, and the Ph.D. degree in Engineering Sciences in 2016 from the University of Trento, Italy. Currently, she is Associate Professor of Universidad de Los Andes. She is currently chair (president) of the Chilean chapter of the IEEE Computational Intelligence Society, and co-organizer of the Chile-WiC, an annual event that seeks to increase the participation of women in computer-related careers. She is also author of the book "Analytics and Big Data: Data science applied to the business world" (ISBN: 978-84-18982-63-7). Her research interests include classification in imbalanced domains, data science in big data applications, computational Intelligence (including machine and deep learning), and data science.