<p>Accurately identifying core technologies is essential for driving technological progress, informing strategic decision-making, and enhancing industrial competitiveness. Existing methods, ranging from expert assessments and quantitative analyses to machine learning techniques, are often hampered by limited adaptability, shallow semantic understanding, and poor time efficiency. To overcome these limitations, we introduce a novel and universally applicable framework that integrates continuous pre-training with multi-task curriculum learning to significantly improve the ability of large language models (LLMs) to identify core technologies and their associated International Patent Classification (IPC) codes across diverse domains. By incorporating extensive, multi-source domain knowledge within a unified continuous pre-training and fine-tuning pipeline, our approach achieves markedly superior semantic comprehension and identification accuracy relative to conventional methods, offering a scalable and effective solution for real-time core technology identification across a wide range of technological domains.</p>