Unleashing the power of 2D CNN with attention and pre-trained embeddings for enhanced online review analysis

Abstract

Online reviews play a significant role in the success of a business. Deep learning models have emerged as crucial tools in this domain, with one-dimensional Convolutional Neural Network (1D CNN) being commonly used. However, this paper proposes a novel approach utilizing a Two-Dimensional Convolutional Neural Network (Att + 2D CNN) with attention mechanism, which effectively captures the dimensionality of the input text, resembling a 2D matrix. To further enhance the model’s performance, we employ pretrained word embeddings, specifically GloVe and Word2Vec. We thoroughly analyze the performance of these embeddings in conjunction with deep learning models. Remarkably, our proposed method, leveraging 2D CNN with attention, consistently achieves superior accuracy when compared to other models, specifically on Amazon Cell Phone reviews and Amazon Kindle reviews datasets, for both balanced and unbalanced natures. By employing this novel methodology, we demonstrate the ability to extract valuable insights from online reviews, enabling businesses to make informed decisions.