A Centered Convolutional Restricted Boltzmann Machine Optimized by Hybrid Atom Search Arithmetic Optimization Algorithm for Sentimental Analysis

• Published in: Neural processing letters Vol. 54; no. 5; pp. 4123 - 4151
• Main Authors: Karthik, E., Sethukarasi, T.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.10.2022; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Arithmetic; Artificial Intelligence; Complex Systems; Computational Intelligence; Computer Science; Computing time; Data mining; Datasets; Decay rate; Deep learning; Machine learning; Mathematical models; Natural language processing; Neural networks; Optimization; Parameters; Randomness; Sentiment analysis; Social networks; Stability analysis; User behavior; United States > US; Sentiment analysis; Hybrid optimization; Centered convolutional restricted Boltzmann machines; Social media; User behavioral analysis
• ISSN: 1370-4621; 1573-773X
• DOI: 10.1007/s11063-022-10797-7

Sentiment analysis uses natural language processing (NLP) to track online conversations and uncover additional information about a subject, business, or theme. Existing machine-learning algorithms are accurate and perform well, but they struggle to reduce computational time and cope with the noisy and high-dimensional feature space of social media data. To resolve these concerns, this paper introduced a Centered Convolutional Restricted Boltzmann Machines (CCRBM), a revolutionary deep learning technique for user behavioral sentimental analysis. The DBN architecture is mainly selected in this work due to its ability to extract in-depth sentimental features, dimensionality reduction, and higher classification accuracy. However, the improper parameter setting can lead to non-convergence, large randomness, and weak generalization capability. To tackle this issue, this work proposes a Hybrid Atom Search Arithmetic Optimization (HASAO) approach, which optimizes DBN parameters such as batch size and decay rate while minimizing DBN issues such as randomness and instability. The performance of the proposed model is analyzed by comparing it with different baseline models and the accuracy value above 90% for the nine datasets proves the efficiency of the proposed technique. When compared to the existing techniques, the proposed methodology offers improved accuracy and speedup capacity.