Purpose <p>The study aims to design and optimize an intelligent education informatization model and recommendation system based on student perception, utilizing a novel Customized Butterfly-Optimized Feed-Forward Neural Network (CBO-FFNN).</p> Design <p>This research employs a hybrid AI-based architecture that combines butterfly optimization with feed-forward neural networks to enhance recommendation accuracy. The system integrates user perception data, borrowing patterns, and course-specific learning outcomes to tailor educational content recommendations. A dataset comprising 42,153 users, 70,000 books, and over 200,000 borrowing records (2021–2024) was analyzed. Additionally, a survey was conducted with 3,516 students enrolled in programming courses, of whom 3,000 responses were considered valid for evaluating satisfaction and system usability.</p> Findings <p>Experimental results demonstrate that the proposed CBO-FFNN model significantly outperforms traditional methods (TF-IDF, CF, and Improved TF-IDF) in accuracy (up to 91.4%), precision (92.1%), recall (92.5%), and F-value (95.7%) across various feature dimensions. Moreover, satisfaction analysis shows over 90% of students found the system useful, relevant to their learning needs, and easy to use.</p> Practical implications <p>The system provides educational institutions with a robust framework for adaptive, perception-aware content delivery, enabling personalized learning and data-driven curriculum refinement.</p> Value <p>This is the first application of a butterfly-optimized neural model tailored to education informatization, integrating both system performance and perceptual feedback to enhance educational recommendation systems.</p>