Turn abandoned patents
into your own filing
into your own filing
This invention describes a computer system that learns to find specific connections (relations) between pieces of information in large amounts of text. It works by taking a collection of text and creating special digital representations of the words that highlight how they relate to specific entities, specifically by converting word distances into continuous vector formats without needing extra linguistic tools. Users provide a few examples of the connections they want to find, which are used to train a neural network, and the system then iteratively identifies the most useful new examples for the user to label, improving its accuracy over time.
Why it matters: Filed before the widespread adoption of large language models (LLMs) and advanced contextual embeddings. Modern LLMs can perform relation extraction with significantly less manual seeding and iterative training, potentially simplifying or outperforming the described neural network training and active learning loop.
AI gives you a few directions you could take this. Pick one, and we check whether your version is different enough to patent, then write the filing.
Reinvent this with AI
