Enhanced Patent Landscape Analysis: A Novel Approach Using Sentence-BERT and Comprehensive Patent Data for Efficient Prior Art Discovery

This study presents a patent search and classification method using Sentence-BERT (SBERT) fine-tuned on Japanese patent data. Using 300 patent records, the model was trained on patent abstracts and International Patent Classification (IPC) data. The fine-tuned model showed improved accuracy in sentence classification and IPC category assignment. Practical applicability was evaluated through similar patent searches using diverse technical terms. Results demonstrated the model's ability to identify relevant patents across various domains, potentially aiding cross-industry innovation. This approach offers efficient similar patent identification and enhanced prior art searches, particularly beneficial for resource-limited enterprises. While promising, further research is needed to validate the method on larger datasets and in real-world scenarios.