Exploring the impact of width in convolutional neural network-based architectures for sentiment analysis

Born from the machine learning (ML) subfield of neural networks (NN), deep learning (DL) has many advantages over other ML algorithms and has become more significant today. As one of the most essential model architectures of DL, the convolutional neural network (CNN) has attracted the attention of many researchers, especially in recent years. Meanwhile, sentiment analysis has become more renowned since the rapid development of various online platforms like blogs, social networks, etc. To study these two heated topics together, this article selects a particular CNN model designed for sentiment analysis and explores its widths potential influence on the result. During the experiment, four CNN models are created based on the same structure but with increasing width. By forwarding the pre-processed datasets to the four models and comparing their performances from different perspectives using different metrics, its concluded that the more expansive the model's width, the better it performs in the training, validation, and testing sections.