From Patent Classification to Technology White Space Identification: A Machine Learning and Language-Model-Based Analysis of USPTO Patent Data

This thesis proposes a machine learning-based analysis of USPTO patent data to classify AI-related patents and identify technology white spaces. The study focuses on patents related to artificial intelligence and data-driven decision systems, using patent text, CPC codes, and metadata to classify them into application-oriented categories such as healthcare, finance, cybersecurity, communication, and general computing. Beyond classification, semantic patent representations and clustering techniques are used to map the patent landscape and detect underexplored areas close to active technological clusters. A language model is then used only as an explanation layer to summarize the identified white spaces in a structured and evidence-based way.