Multi-Class Classification of Customer Complaints Using Convolutional LSTM and Neural Network Models

Abstract

This paper provided multi-class classification models for customer complaints based on neural network architectures. It considered one-dimensional convolutional models corresponding to one, two, and three layers, and also included Long Short-Term Memory and Convolutional Long Short-Term Memory models. The models were assessed through performance indicators such as accuracy, area under the curve, precision, and recall. The findings indicated that Convolutional Long Short-Term Memory exhibited superior performance, achieving 88.06% accuracy and 0.92 area under the curve, respectively. This outperformed one convolutional layer models, Long Short-Term Memory, two convolutional layers, and three convolutional layers models, whose accuracy rates were 86.48%, 84.02%, 83.87%, and 74.71%, respectively. Moreover, Convolutional Long Short-Term Memory also surpassed other models in terms of precision and recall, achieving 0.89 and 0.86, respectively. The study demonstrated that it is superior to simple models to apply a hybrid Convolutional Long Short-Term Memory to classifying customer complaints. This technique combines effectively the advantages of convolutional and recurrent layers, facilitating an ability to learn both local and long-term features. The findings underscored the significance of appropriating optimal neural network architectures for complex text classification problems and also made an original contribution to research on consumer feedback.