Phoenix–Chameleon Optimization for Feature Selection in Sentiment Analysis
K Fathima, A. R. Mohamed Shanavas · Indian Journal of Science and Technology · 2025
Objective: To develop a hybrid optimization algorithm for robust feature selection in sentiment analysis. Methods: A Two-Phase optimization procedure is proposed, Phoenix-Chameleon Optimization Algorithm (PCOA), which has the entropy-driven exploration phase and density-related exploitation phase combined with each other. Phoenix phase supports global search based on rebirth, whereas Chameleon phase supports local fine-tuning based on feature co-occurrence density and most recent gains. TF-IDF and RoBERTa embeddings are represented by feature vectors that are reduced in terms of PCOA and tested by BiGRU classes. The proposed work performed experiments on four benchmark corpora, MR, CR, IMDB and SemEval 2013. Findings: Accuracy of PCOA is up to 96.98% on IMDB, but on all the datasets performance increases (3%-6%). Using the F1-scores and training efficiency as a metric, PCOA attains better results compared to Chi-Square, Recursive Feature Elimination (RFE) and Particle Swarm Optimization (PSO) in both lexical and contextual embeddings. Novelty: PCOA introduces a new entropy-based phase-switching mechanism between Phoenix and Chameleon behaviors, creating a domain-robust, dynamic feature selection framework. Keywords: Feature selection, Sentiment analysis, Phoenix optimization, Chameleon optimization, Deep learning