Since the 2010s, advances in machine learning (ML), particularly in neural networks and deep learning, have fueled artificial intelligence (AI) growth. The release of GPT 4.0, a multimodal model processing text and image inputs, spotlighted generative artificial intelligence (GenAI). The GenAI market is expected to grow from US$36 billion in 2024 to US$356 billion by 2030. Although AI literature is extensive, a systematic guideline, a comparison of machine learning algorithms for identifying these technologies, and an overview of GenAI technologies in the US are missing. This study addresses these gaps by exploring the approach and the implementation of supervised machine learning models to identify GenAI technologies in patents. Using traditional algorithms and transformer-based neural networks, over 2.4 million patents from the USPTO database are classified. A framework for training, evaluating, and optimizing ML models is presented, with XLNet proving the most accurate and robust classifier. Findings reveal significant growth in GenAI patents, concentrated in industries and regions like California, USA. This research provides theoretical insights into GenAI innovation, practical guidance for replicating the ML methodology to other contexts, and identifies future research opportunities such as full-text patent analysis. It offers a fundamental tool for advancing the study of GenAI-driven innovation.
