An LLM-Augmented ML Framework for Cross-Domain Sentiment Analysis
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This paper presents a novel LLM-augmented machine learning framework for cross-domain sentiment analysis that combines traditional ML approaches with large language model assistance.The proposed framework integrates TF-IDF feature extraction, ensemble classification methods (SVM, Random Forest, Gradient Boosting) and dimensionality reduction techniques (LSA, LDA) to achieve competitive performance while maintaining superior computational efficiency and interpretability.Evaluated across three heterogeneous domains-electronics, food & beverage and apparel reviews-the framework achieves 83.7% accuracy with only 2.7% cross-domain degradation.Key innovations include transparent LLM integration for research augmentation, weighted ensemble voting mechanisms and systematic hyperparameter optimization via GridSearchCV.The framework demonstrates practical viability for resource constrained environments, achieving 20-50× faster inference (2.1ms vs 45-120ms) and 8-10× smaller model size compared to deep learning alternatives, while maintaining explainability crucial for regulated domains.
Fields of Study & Subjects
Sentiment Analysis and Opinion Mining
Topic Modeling
Text and Document Classification Technologies
Publication details
- Journal / source
- Journal of Artificial Intelligence Machine Learning and Data Science
- Volume
- 9
- Issue
- 1
- Pages
- 3286–3291
- Publication date
- 18 March 2026
- Language
- EN