Knowledge Graph Embedding Patents: Leaders, Trends & White Space 2026
Filing growth compares 2021 (2,124 records) with 2024 (2,825) — a three-year span. 2024 is the most recent year we treat as complete: publication lags filing by roughly 18 months, so 2025 onwards are still filling in and any growth rate that ends there would understate the field. Top-5 share is the combined record count of the five largest assignees divided by all 68,559 records in scope (CR5), not by the ranked leaders only.
What the knowledge graph embedding patent record shows
Knowledge graph embedding turns entities and relations into vector representations that support link prediction, reasoning and downstream machine learning tasks. The patent record for this field is large and still expanding: 68,559 published records sit within scope between 2015 and the 2026 data cut-off, with filing activity climbing through 2024 before the most recent years understate themselves due to publication lag. The technology composition leans heavily on general data-processing and AI-model classes rather than a single narrow art unit, which tells its own story about how broadly this technique has been absorbed into adjacent systems.
No single assignee controls the field. The ranked leader accounts for a meaningful but far from dominant share of records, and the combined share held by the top ranked filers stays in the single digits to low teens when measured against the full record count. That pattern — a recognisable leader group sitting atop a long tail of single- and few-filing entrants — is typical of a technique that has moved from research labs into broad commercial infrastructure faster than any one company could claim it.
Let an AI agent run this analysis on your own technology
Pick a task. Every answer cites the patents behind it.
Filing trend and IPC composition
The figures below track publication volume by year and the IPC subclasses that carry the most knowledge graph embedding filings. Because a single record can carry several IPC codes, subclass shares are read against the full record count and will sum to more than 100%.
Growth phase, not a plateau
Filings rose from 594 in 2017 to a peak of 2,825 in 2024, including a documented +33% climb from 2,124 (2021) to 2,825 (2024). The 2025 and 2026 figures are lower only because publication typically lags filing by about 18 months — they are not evidence of a slowdown.
Concentration in general data processing and AI classes
G06F (electric digital data processing) covers 16.0% of the 68,559 records in scope and G06N (AI-model computing) covers 11.1%, together dwarfing the narrower application classes such as G06Q, H04L, G06V, G06T, G10L and G06K, each in the 1.9%–3.9% range. That gap signals that embedding techniques are being claimed mostly at the infrastructure and model layer rather than tied to a single downstream application.
Shares are the percentage of the 68,559 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on Knowledge Graphs & Graph Computing: Knowledge Graph Embedding Patent Landscape with Eureka
This page is one run against one query. Ask Eureka your own question about knowledge graphs & graph computing: knowledge graph embedding patent landscape and every answer comes back with the patent numbers behind it.
Try EurekaA representative knowledge graph embedding patent
Method for inductive knowledge graph embedding using relation graphs and system thereof
Disclosed is an inductive knowledge graph embedding method and system through a relation graph. The method trains a graph neural network-based knowledge graph embedding model using a knowledge graph and a relation graph generated from it, then performs link prediction for a knowledge graph that includes a new relation and a new entity through the trained model.Published 2024-12-05 — representative of a shift toward inductive embedding methods that handle entities and relations unseen at training time.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US6850252B1 | Intelligent electronic appliance system and method | 4,059 |
| 2 | US20030229900A1 | Method and apparatus for browsing using multiple coordinated device sets | 3,475 |
| 3 | US20090254572A1 | Digital information infrastructure and method | 2,827 |
| 4 | US6571282B1 | Block-based communication in a communication services patterns environment | 2,353 |
| 5 | US20110258049A1 | Integrated Advertising System | 2,187 |
| 6 | US20040031058A1 | Method and apparatus for browsing using alternative linkbases | 1,955 |
| 7 | US20070087756A1 | Multifactorial optimization system and method | 1,866 |
| 8 | US20100250497A1 | Electromagnetic pulse (EMP) hardened information infrastructure with extractor, cloud dispersal, secure stora… | 1,777 |
| 9 | US20080177994A1 | System and method for improving the efficiency, comfort, and/or reliability in Operating Systems, such as for… | 1,735 |
| 10 | US20190200844A1 | Method of HUB communication, processing, storage and display | 1,720 |
Citation counts reflect an older, heavily-cited cohort within the searched corpus and should be read as a signal of influence on the field's early architecture rather than of current commercial importance.
Each row carries its publication number; clicking a row searches Eureka by that number.
Put your own technology through the same analysis
Eureka on the web
When you want the answer in the next five minutes.
The agent works the prompt against patents and technical literature, citing every source.
Run your analysis now →MCP server & REST API
When it has to run inside your own pipeline.
Patent search, landscape analysis and assignee resolution as MCP tools. Drop them into any agent framework, or call REST directly.
Browse MCP servers →What the numbers mean for filing strategy
Three patterns stand out once family counts, assignee concentration and technology composition are read together.
A long tail, not a duopoly
The ranked leader holds 1,976 records, but the top 5 combined reach only 9.3% and the top 10 only 13.0% of all records in scope. That leaves the large majority of filings distributed across a wide field of entrants — a structure where a new filer is unlikely to face a single blocking incumbent but also unlikely to find an empty field.
Still climbing through the last complete year
Filings grew from 2,124 in 2021 to a peak of 2,825 in 2024, the most recent year with reliable publication coverage. Because publication lags filing by roughly 18 months, the apparent dip in 2025-2026 reflects incomplete data, not reduced inventive activity.
Claims sit at the infrastructure layer
G06F and G06N together account for the largest shares of the 68,559 records, well ahead of application-specific classes like G06Q, H04L or G10L, each under 4%. Filers are claiming embedding methods and model architectures more than they are claiming specific end-use applications.
US-centred, with PCT and EPO as secondary routes
The United States receives by far the largest share of filings at 13,788 records, with WIPO/PCT (2,567) and the EPO (2,207) as the next largest routes, followed by Australia, Canada and the UK at smaller volumes. That distribution favours a US-first filing strategy with PCT as the standard route to broader coverage.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to knowledge graphs & graph computing: knowledge graph embedding patent landscape, with the prior art for and against each one.
| Assignee | Co-assignee | Shared families |
|---|---|---|
| International Business Machines Corporation (IBM) | IBM (China) Co., Ltd. | 26 |
| International Business Machines Corporation (IBM) | Rensselaer Polytechnic Institute | 17 |
| International Business Machines Corporation (IBM) | IBM UNITED KINGDOM LTD INTELLECTUAL PROPERTY DEPARTMENT | 15 |
| Microsoft Technology Licensing, LLC | WU XIANCHAO | 14 |
| International Business Machines Corporation (IBM) | Massachusetts Institute of Technology | 8 |
| Digimarc Corporation | ACOUSTIC INFORMATION PROCESSING LAB | 8 |
| International Business Machines Corporation (IBM) | The Board of Trustees of the University of Illinois | 6 |
| International Business Machines Corporation (IBM) | IBM ISRAEL SCI & TECH LTD | 6 |
Ten co-assignee pairs appear in the data, the strongest linking a single large corporate group with its regional subsidiary and academic partners — a pattern consistent with joint research filings rather than broad industry consolidation.
Where to take this analysis
The figures above establish scale and shape. Turning them into a filing or freedom-to-operate decision requires drilling into specific claim sets and named assignees.
Map claim overlap against a specific architecture
Run a claim-level comparison against the inductive and transductive embedding methods already crowding G06F and G06N before drafting new claims in those classes.
Explore claims in Eureka →Track the long tail of single-filing entrants
With the top 10 holding only 13.0% of records, most of the field's inventive activity sits outside the ranked leaders — worth monitoring for early acquisition or licensing targets.
Set up assignee tracking in Eureka →Watch application classes for emerging pressure
G06V, G06T and G10L each sit under 3% of records today; a jump in any one signals embedding techniques moving into a new commercial application faster than the broader trend.
Monitor IPC shifts in Eureka →Common questions about knowledge graph embedding patents
There are 68,559 published records in scope for knowledge graph embedding between 2015 and the 2026 data cut-off. This figure counts records, not distinct inventions, since a single technology can generate multiple family members across jurisdictions. Filing volume grew steadily through the period, peaking at 2,825 in 2024, the most recent year with complete publication data.
The ranked leader holds 1,976 records, well ahead of the rest of the field, but no single company controls the space: the top 5 assignees combined account for only 9.3% of all 68,559 records, and the top 10 for 13.0%. This is a long-tail field with a wide base of smaller filers rather than a concentrated duopoly. Ranking positions are based on the 100 assignees the data endpoint returns as the full ranking, not a curated top-50 or top-100 list.
Yes, through 2024. Filings rose from 2,124 in 2021 to 2,825 in 2024, a 33% increase over that span. Figures for 2025 and 2026 appear lower, but that reflects the roughly 18-month lag between filing and publication rather than an actual decline in inventive activity, so those years should not yet be read as a slowdown.
The largest concentration sits in G06F (electric digital data processing), covering 16.0% of the 68,559 records, and G06N (computing based on AI models), covering 11.1%. Narrower application classes such as G06Q (business data processing), H04L (digital transmission), G06V (image and video recognition), G06T (image data processing), G10L (speech and audio) and G06K (data recognition) each account for under 4% of records. Because records can carry multiple IPC codes, these shares sum to more than 100% and should not be added together to describe field size.
The application-layer classes are comparatively underclaimed relative to the core infrastructure classes: G06V, G06T, G10L and G06K each sit at or below 2.5% of the 68,559 records, versus 16.0% for G06F and 11.1% for G06N. That gap suggests embedding methods tied to specific application domains like speech or video recognition carry less claim density than the general model and data-processing layer. Combined with a long tail of assignees rather than a concentrated leadership group, a new entrant focusing on a specific application integration rather than a general embedding architecture is less likely to run into dense prior art.
Research Knowledge Graphs & Graph Computing: Knowledge Graph Embedding Patent Landscape in depth with Eureka
Go past this page: query the whole knowledge graphs & graph computing: knowledge graph embedding patent landscape corpus yourself, in your own scope.
Every answer comes back with patent numbers you can open.
Disclaimer. This page is generated from Patsnap Eureka data drawn from a limited snapshot of global patent and scientific-literature records, and is provided for general information and reference only.
Patent data carries inherent limitations: recent filings (typically the most recent 18–24 months) are under-counted due to standard publication lag; counts may be reported at either a patent-family or a patent-record basis and are not always directly comparable; classification, applicant-name, and citation data may contain errors, duplicates, or omissions; and the underlying search query defines and constrains the scope shown. As a result, the analysis may be incomplete or inaccurate and may not reflect the full technology landscape.
Nothing on this page constitutes an exhaustive prior-art, novelty, freedom-to-operate, or validity search, nor does it constitute legal, financial, investment, or professional advice, and it should not be relied upon as such. Any patent, commercial, or strategic decision should be verified independently and reviewed with qualified patent, legal, and domain professionals. Patsnap makes no warranties, express or implied, as to the accuracy, completeness, or fitness for any particular purpose of the information presented.
Machine translation. Assignee and organisation names originally recorded in Chinese, Japanese or Korean have been rendered into English by an AI translation step so that the tables stay readable. These renderings are best-effort and may not match a company’s registered English name; the original name is what the underlying patent record carries, and it is what any Eureka query launched from this page uses.