Multi-Label Classification in Text Data: An Examination on Innovative Technologies

Abstract

This study focuses on examining the effectiveness of multi-label text classification methods in the field of innovative technology. Experimental results using four different transformer models such as bert-base-cased, albert-base-v2, xlm-roberta-base and bart-base reveal in detail the performance of these models on performance measures such as accuracy, F1 score and processing time. The study highlights another critical factor in model selection, the balance between accuracy and processing time, providing researchers and practitioners with valuable information on the selection of models suitable for specific use scenarios. Key findings of the study include bert-base-cased standing out with high accuracy and F1 score, while xlm-roberta-base stands out with competitive performance and improved processing efficiency. These results evaluate the success of transformer models in multi-label text classification tasks and reveal the factors to be considered in choices in this field. The study provides a framework for future research, guiding future studies on topics such as model discovery, development of fine-tuning strategies, and model interpretability. © 2024 IEEE.