Structured Technology Intelligence from Published Patents.
Polyrek explores how fragmented technical disclosures in patents can be transformed into structured, reusable technology knowledge and whether those structures can help us understand how technologies form, evolve and recombine.
Patent data is abundant. Technology knowledge is still fragmented.
Search, classification and summarization are increasingly accessible. The harder problem is turning document-level disclosures into persistent technical concepts that engineers and researchers can reason about.
Documents versus technologies
A single patent may disclose multiple technical features, while the same technical concept may appear across unrelated documents and assignees.
Similarity versus technical meaning
Textual similarity alone does not reliably distinguish common problems, alternative solutions, structural differences and technical effects.
Knowledge needs structure
The project investigates whether patent disclosures can be decomposed and reorganized into reusable units suitable for clustering and longitudinal analysis.
Decoding technology from patents.
RQ1 · Technology Concept Formation
Can PSCE-based centroid representations form coherent, distinguishable and stable technology clusters from patent technical features?
Current work examines cluster formation, threshold behavior, pool and official cluster dynamics, and stability as new patent evidence is introduced.
View research planWhat exists today.
These are working capabilities and preliminary research artifacts, not claims that the long-term hypothesis has been validated. View lastest research
Structured technical features
Patent disclosures are decomposed into feature-level representations using Problem, Solution, Component or Configuration, and Effect.
Centroid-based concept formation
Feature centroids are used to form candidate pool clusters and stabilized official clusters under configurable assignment and guard conditions.
Latest Records & Notes
Research Records
Notes
Commercial patent intelligence platforms provide technology landscape functions that organize large patent collections into clusters or visual maps. Although the detailed algorithms used by these systems are generally proprietary, some vendors publicly describe the major steps involved in constructing such landscapes.
Open to research and validation partners.
Current interests include domain validation, patent analytics research, external datasets and industrial evaluation of technology-level patent intelligence.
About the project