Deepsentimodels: A Novel Hybrid Deep Learning Model for an Effective Analysis of Ensembled Sentiments in E-Commerce and S-Commerce Platforms. Issue 4 (19th May 2023)

• Record Type: Journal Article
• Authors: Venkatesan, R.; Sabari, A.
• Is Part Of: Cybernetics and systems. Volume 54:Issue 4(2023)
• Page Start: 526
• Page End: 549
• Publication Date: 2023-05-19
• Subjects: Bi-convolutional neural networks -- emotional catalysts -- optimized long short term memory -- social networks -- spotted hyena; Cybernetics -- Periodicals; System theory -- Periodicals; 003.5
• DOI: 10.1080/01969722.2022.2148510
• Languages: English
• ISSNs: 0196-9722

Abstract

Abstract: Social networks and media have gradually grabbed significant time from people's lives to share and communicate information. Moreover, these social platforms also act as an emotional catalyst for expressing their feeling and views on different products, movies, and even national policies. Thus, it helps to understand people's opinions and predict their behaviors in social networks, where it helps the entrepreneur improve their products or services. It is considered the most efficient method to understand the customer's view on their product better. Since entrepreneurs are increasing exponentially, an intelligent system is required to understand people's views on their products. Also, people use informal language to express their sentiments; it has become a daunting task to extract the sentiments that reflect the people's attitude and feedback about the product. Therefore, this article suggests a novel hybrid deep learning system, consisting of bi-convolutional networks (B-CNN) and spotted hyena optimized long short term memory (SHOLSTM), for improved understanding of people's feelings and the construction of intelligent recommendation systems for entrepreneurs. Many experiments have been conducted utilizing various datasets compared to other current hybrid deep learning methods, including long short-term memory (LSTM), convolutional neural networks (CNN), BIGRU, and attention classifiers. Accuracy is achieved as 99.4%, 99% precision, 99.2% recall, 99.2% F1-score is achieved, which suits these algorithms best for implementing the intelligent recommendation systems based on the different ensembled sentiments.