Optimization-enabled deep learning for sentiment rating prediction using review data

At present, websites, such as Amazon and Yelp, permit consumers to submit reviews for different businesses, services, and products. In recent times, the customer’s shopping decision is highly influenced by the online reviews posted by the customers using star ratings, as well as, texts. Generally, customers believe that ratings are consistent with the reviews provided, but it may not be in the case of intermediate ratings. Conversely, the reviews posted by the customers will be in an unstructured format, which makes it highly complicated in understanding and analysis of the review, thereby requiring efficient approaches to estimate ratings considering the sentiment of the customer. In this paper, a novel sentiment rating prediction approach is devised by utilizing a deep learning network with hybrid optimization. Here, Random Multimodel Deep Learning (RMDL) network is used to estimate the sentiment rating, where the RMDL is trained using the devised Honey-based Exponential Poor Rich Optimization (HEPRO) algorithm. Moreover, the presented HEPRO-RMDL is examined for its effectiveness based on various metrics, like True Positive Rate (TPR), True Negative Rate (TNR), and accuracy is observed to have achieved high values of accuracy, TNR, and TPR at 0.935, 0.925, and 0.956, respectively.