Differential Evolution-Initialized Cuckoo Search for AI-Based Feature Selection in CoWIN Twitter Sentiment Analysis

Abstract Content in social media is hard to be analyzed by sentiment as there is increasing amount of social media content related to health, and the dimensionality and computation required to handle the redundant and irrelevant features in the content representation. In this research, a feature selection method based on DE-Initialized Hybrid Cuckoo Search Algorithm (HCSA) is suggested for Twitter data to be used in health-related sentiment analysis. The proposed method is based on the Cuckoo Search (CS) optimization algorithm where a Population Initialization Strategy (DE) is introduced. The approach is applied to the features as well as the sentiment classification, and the features are represented by the Term Frequency–Inverse Document Frequency (TF-IDF) matrix and the sentiment classification is provided by Support Vector Machine (SVM) classification algorithm taking the CoWIN Twitter sentiment data set as the data set. The experimental setup is identical to the Cuckoo Search (CS), Differential Evolution (DE), Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) to make comparisons. From the results, it is seen that the number of features is reduced by 48.40% from 500 to 258 dimensions with no reduction in the classification accuracy for the HCSA. The highest accuracy and F1 scores that can be achieved from the observed HCSA are 82.90% and 73.51%, respectively, with the highest macro-precision and macro-recall scores of 85.55% and 73.05%, respectively. These results suggest that the proposed DE-initialized HCSA can effectively compress the text's dimensionality in the meantime retaining useful information for sentiment classification. The study found that hybrid metaheuristic feature selection could be one of the most effective feature selection methods to extract the information about health from social media text.