<p>Particle Swarm Optimization-Convolutional Neural Network (PSO-CNN): An Optimized Deep Learning Model for Personality Recognition From Social Media Profiles</p>

Abstract

Recognition of users' accounts on social media platforms remains challenging, particularly when profile images are hidden or users maintain multiple accounts. This paper proposes and validates a novel Particle Swarm Optimization-Convolutional Neural Network (PSO-CNN) hybrid model that fundamentally addresses the critical challenge of hyperparameter optimization in automated personality recognition from social media. Unlike existing approaches that rely on manual trial-and-error configuration, the PSO-CNN framework uniquely automates the discovery of optimal architectural parameters by integrating the global search capabilities of PSO with the feature extraction power of CNNs. This automation eliminates subjective design choices, uncovers non-intuitive parameter combinations that human experts might overlook, and ensures consistent performance across diverse personality types without manual adjustment. It automatically selects optimal learning rates, filter configurations, and regularization settings for classifying users according to the Myers-Briggs Type Indicator framework. Evaluated on a Twitter dataset of 8,328 users, the PSO-CNN-optimized model achieved significant performance with 84.5% test accuracy and 98.2% F1-score, representing improvements of 14.6% and 21.8%, respectively, over a standard CNN baseline. The optimized model demonstrated robust performance across all 16 personality types, maintaining high precision (98.9%) and recall (99.8%) with a low false positive rate of 0.3% while converging to optimal parameters within 65 PSO iterations. These findings empirically validate that metaheuristic optimization substantially enhances deep learning performance for psychometric applications, offering an efficient, scalable solution for personality recognition with practical implications for personalized systems, human-computer interaction, and computational social science. The study establishes PSO-CNN as an effective paradigm for advancing automated personality assessment while providing a foundation for future research into multimodal integration and cross-platform generalization.