A Review of Metaheuristic Algorithms Applications in Text Sentiment Analysis

Text sentiment analysis, as one of the main branches of natural language processing, focuses on identifying and classifying opinions and emotions in various texts, with wide applications in fields such as marketing, healthcare, and policymaking. With the growth of social media and the massive production of textual data, high dimensionality and noisy features have become fundamental challenges. Feature selection emerges as an effective solution for dimensionality reduction and improving model performance, and metaheuristic algorithms have attracted researchers' attention due to their ability to search complex spaces. This review paper examines reputable studies to classify metaheuristic approaches in text sentiment analysis. The algorithms studied include improved versions (MBHOA, MEHO), hybrid methods (Aẞ-BGWHHO, GNN-DE), ensemble approaches (MBPSO-BDGWO), combination with deep learning (GNN-MEHO), and hyperparameter optimization (Optuna). Results indicate that hybrid and ensemble methods, by creating a better balance between exploration and exploitation, achieve higher stability and accuracy compared to single algorithms. Combining metaheuristic algorithms with graph neural networks reduces premature convergence and improves classification accuracy. Hyperparameter optimization also has an impact beyond feature selection itself. Key challenges include overfitting, domain adaptation, sarcasm detection, data sparsity, computational complexity, and ethical issues. Future directions include adaptive feature selection, multimodal analysis, graph neural networks, self-supervised learning, simultaneous optimization, and ethical considerations.

کلیدواژه ها:

Metaheuristic Algorithms ، Text Sentiment Analysis ، Feature Selection ، Deep Learning