Efficient AI and Deep Learning for Sentiment Analysis: Optimized Hybrid MLP for Low-Computational Setups
Year: 2025
Type: article
Abstract: – This paper presents an Optimized Hybrid Multi-Layer Perceptron (O-HMLP) for sentiment analysis, designed specifically for low-computational environments where access to powerful hardware is limited. The model integrates handcrafted linguistic features, dimensionality reduction, clustering-based signals, and a lightweight deep feedforward neural architecture. Experiments conducted on the IMDb dataset demonstrate that the O-HMLP achieves 70.2% accuracy in only 19.5 seconds of training, representing a significant improvement over the baseline Hybrid MLP (51.0% accuracy, 3.8s).. Compared to SVM and Naïve Bayes, the O-HMLP provides a competitive balance between accuracy and efficiency, though it still lags behind state-of-the-art models such as BERT and Transformers. Importantly, the architecture remains experimental and untuned, meaning results may vary; however, its extensibility suggests a higher likelihood of improvement with proper optimization. The long-term research goal is to achieve 85%+ accuracy while maintaining training time under 25 seconds, offering a practical solution for inclusive NLP in resource-constrained scenarios.
Author: Perez, John David L.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7083426176
FWCI:
Cites:
Cited by:
Related to: