The Deep Learning Techniques for Analyzing Sentiments from Customer Reviews in Corporate Branding

Marketers have been presented with unparalleled possibilities as a result of the growing influence of social media. One strategy that appears to be gaining traction is the use of brand-owned social media to raise awareness, foster relationships with consumers, and increase interaction between the brand and its target audience. This research presents an innovative approach based on deep learning (DL) to understand customers' emotions through their reviews of the products and understand trust and loyalty in the brand. This paper proposes a solution that uses a pre-trained transformer model and develops a novel customized hybrid model using convolutional neural networks and long short-term memory. The experiments are performed on the open-source Kaggle dataset, and the data is modified to extract only the review text and score to perform sentimental analysis. The text is pre-processed to extract the tokens and generate the embeddings to be trained on the DL models. Among the two proposed models, the highest-performing model in terms of all the evaluation metrics is the BERT model, which achieves the highest accuracy of 92% with good generalization. However, the time required for training and validation is relatively high compared to the hybrid model. Considering the accuracy and time tradeoff, the better model is BERT, but when only the computation efficiency needs to be considered, then the hybrid model is good enough. Thus, implementation of the proposed model in real-time can be beneficial to understanding the sentiments of the reviewer for the product, which helps to enhance the services.