Project Overview
The USPTO Forensics project established a data-driven framework to quantify the "average" technological focus of patent Art Units and measure how individual patents or examiners deviate from those norms. By bridging the gap between patent analytics and organizational policy, the system enables stakeholders to identify emerging subfields and make evidence-based decisions regarding resource allocation.
Key Features
- Semantic Analysis through PatentsBERTa to convert complex patent text into 768-dimensional mathematical vectors
- Centroid Profiling to create a statistical fingerprint of an Art Unit's core technological domain
- Deviation Scoring to identify outlier patents and score examiners based on technological consistency
- Visual Exploration using UMAP dimensionality reduction to project high-dimensional data in a digestible format for stakeholders
Meet the Team
Tech Stack
- Frontend: Nuxt.js/Vue 3/Tailwind CSS
- Backend: Node.js/Express and FastAPI (Python)
- Databases: MongoDB (Metadata) and Pinecone (Vector Embeddings)
- Model: PatentsBERTa