In this work introduces a data-centric weakly supervised deep learning framework for domain-specific patent identification, addressing the challenge of limited annotated data in specialized domains. It will help researchers automatically identify patents within a specific domain.
The identification of patents in a specific technological area is a important task during the early phases of an R&D project. It is crucial for scientists and researchers to identify patents relevant to their fields in order to maximize the value of their intellectual property (IP) and anticipate emerging challenges. For industries, this task is essential, as it allows to gain valuable insights into technological advancements, market competition, and future opportunities for innovation. In this work, we introduce a data-centric weakly supervised deep learning framework for domain-specific patent identification. Our approach follows a data-centric AI methodology, generating a training dataset by leveraging multiple sources and features such as linguistic patterns, heuristics, and domain-specific knowledge. We integrate several labeling functions and apply a generative model for producing high-quality labels for downstream training. Based on this, we fine-tune a pre-trained language model for the patent identification task by adapting its hyperparameters and training it on the training dataset, enabling the model to learn domain-specific patterns and improve classification accuracy. We evaluate the model on unseen test data by using common performance metrics and compare the achieved results against an in-context learning approach.
References
@article{SOFEAN,
author = {Mustafa Sofean},
title = {Identification of domain-relevant patents via weakly supervised deep learning},
journal = {World Patent Information},
volume = {84},
pages = {102434},
year = {2026},
issn = {0172-2190},
doi = {https://doi.org/10.1016/j.wpi.2026.102434},
url = {https://www.sciencedirect.com/science/article/pii/S0172219026000098},
}
This project is licensed under the Apache License 2.0. See the LICENSE file for details.