In a groundbreaking advancement in artificial intelligence and innovation analysis, scientists from the Seoul National University of Science and Technology (SeoulTech) have developed an AI-powered Patent Abstract Generator capable of translating patent vacancies into human-readable text.
This pioneering technology is designed to accelerate the discovery of emerging technologies and optimize R&D strategies for companies and researchers worldwide.
Patent Abstract Generator Simplifies Identification of Technological Opportunities
Identifying technology gaps within patent maps – a key process for discovering untapped innovation opportunities – has long been a challenge.
Patent maps are visual representations of technological landscapes created using dimensionality reduction techniques. However, interpreting these maps to pinpoint and understand vacant areas has remained difficult.
To address this, a team led by Professor Hakyeon Lee, Department of Industrial Engineering, SeoulTech, introduced a new machine learning approach based on text-embedding inversion.
This approach reverses complex data embeddings back into human-readable language, effectively translating abstract patent vacancies into detailed and understandable technological descriptions.
Five-Step AI Framework for Translating Patent Vacancies
The Patent Abstract Generator operates through five major stages:
- Transformation of patent abstracts into high-dimensional vectors via text embedding.
- Training of an autoencoder to project embeddings into 2D space for bidirectional mapping.
- Creation of grid-based patent maps using kernel density estimation.
- Detection of vacant cells and coordinates representing patent gaps.
- Reconstruction of these coordinates into readable patent abstracts via a vec2text decoding system.
This methodology was detailed in the journal Advanced Engineering Informatics (Volume 68, Part B, November 2025), with findings first made available online on July 28, 2025.
From Patent Vacancies to Readable Innovations
According to Professor Lee, “The most revolutionary aspect of our research is its ability to translate abstract patent vacancies into concrete, human-readable technology descriptions. Unlike previous systems that merely identified empty spaces on patent maps, our AI model generates detailed abstracts describing the potential technology that could fill those gaps.”
The researchers demonstrated the system’s efficacy through a case study on LiDAR technology using 17,616 patents.
The Patent Abstract Generator successfully identified patent gaps and produced descriptive abstracts, underscoring its potential as a transformative tool for technology forecasting and innovation management.
Democratizing Global Innovation
Professor Lee further noted that this technology could “fundamentally democratize innovation forecasting.” The Patent Abstract Generator could enable startups, research institutions, and policymakers to identify breakthrough opportunities previously accessible only to large corporations with advanced R&D resources.
The SeoulTech research team is already working on extending this system to automatically generate full research proposals and patent documentation – potentially creating a seamless AI-driven innovation ecosystem from discovery to development.
