An Empirical Study of Loss Functions for Aspect Category Detection in Imbalanced Data Scenario

One of the key challenges in the Aspect Category Detection task is dealing with class imbalance in the data, which can significantly affect the performance of machine learning models. In this study, different loss functions have been analyzed to tackle the problem of class imbalance for Aspect Category Detection from textual data. A comprehensive comparative analysis of loss functions, including Binary Cross-Entropy (BCE), Weighted Binary Cross-Entropy (WBCE), Class-Balanced Cross-Entropy (CBCE), and Focal Loss have been conducted, across Bi-LSTM and Bi-GRU architectures. The performance of the models has been evaluated on two imbalanced datasets using key metrics such as precision, recall, macro <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$f$</tex> 1-score, accuracy and MCC. The results indicate that focal loss consistently outperforms other loss functions, particularly when used with the Bi-LSTM model, achieving the highest scores and demonstrating superior handling of class imbalance issue.