A Transformer-Based Approach to Tag Prediction in Competitive Programming Problems

Artificial intelligence reasoning ability is crucial for enhancing problem-solving and decision-making algorithms. Competitive programming problems involving logical reasoning and analytical thinking represent a perfect domain for testing the reasoning capability of AI. Automated tag prediction for competitive programming problems is examined in this study as a step toward reasoning-aware AI. 3,133 single-tag problems were chosen from an initial set of 8,000 problems to provide clear learning. These issues were labeled into seven separate classes so that the task could be considered as a single-label classification problem. Code Bidirectional Encoder Representations from Transformers, a transformer-based pre-trained model, was fine-tuned end-to-end by modifying all the layers to learn the specifics of this task. For class imbalance, class weights were added during training. The model was trained on a stratified train-test split and assessed according to accuracy and confusion matrices. The optimized Code Bidirectional Encoder Representations from Transformers achieved a maximum accuracy of 61%, demonstrating its promise for automatic tag prediction and indicating areas for further improvement. This study emphasizes the importance of competitive programming in creating reasoning-based AI systems and lays the foundation for future research in this area.