Volume 23, Issue 60, July - December, 2026

Optimized Hierarchical Auto- Associative Polynomial Convolutional Transformer Network for Accurate Sentiment and Rating Prediction from Customer Reviews

Dinesh MG¹, Manisha G², Gunanandhini S³, Anish TP⁴, Thilagam T^5♦, Siva Subramanian R⁶

¹Dept of AI&DS, Easa College of Engineering and Technology, India
²Dept of CSE, R.M.D. College of Engineering and Technology, India
³Dept of CSE, S.A. Engineering College, India
⁴Dept of CSE, R.M.K. College of Engineering and Technology, India
⁵Dept of CSE, Vel Tech Rangarajan Dr. Sangunthala R&D Institute of Science and Technology, Chennai, India
⁶Dept of CSE, R.M Institute of Science and Technology, India

^♦Corresponding Author
Thilagam T, Dept of CSE, Vel Tech Rangarajan Dr.Sangunthala R&D Institute of Science and Technology, Chennai, India

ABSTRACT

The sentiment analysis and star rating prediction are essential to understand the consumer feedback from online reviews. Traditional approaches may not be sufficient to represent both nonlinear feature interactions and contextual relationships. To overcome these drawbacks, in this work, a Hierarchical Auto- Associative Polynomial Convolutional Network-based Geometric Algebra Transformer with Duck Swarm Optimization (HAPCNbGAT-DSO) is proposed. The framework is based on a hierarchical polynomial convolution for nonlinear feature learning, a Geometric Algebra Transformer (GATr) for relational contextual modelling, and Duck Swarm Optimization (DSO) for global hyperparameter search. After systematic pre-processing and feature representation, the model is tested on the Yelp Restaurant Review Dataset (YRRD) and Restaurant Review Dataset (RRD). The experimental results show the accuracy of 0.997 for RRD and 0.998 for YRRD, and the corresponding loss is 0.03 and 0.02 respectively. The study also assesses interpretability, cross domain generalization, computational efficiency, optimizer sensitivity, statistical significance and component-wise ablation.

Keywords: Duck swarm optimization, Geometric Algebra Transformer, Hierarchical Auto-Associative Polynomial Convolutional Network, natural language processing, Shapley Additive Explanations

Indian Journal of Engineering, 2026, 23(60), e16ije1724

PDF

Published: 12 October 2026