Fastest
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 →https://www.patsnap.com/resources/blog/rd-blog/exploration-and-bandit-methods-patent-landscape/ · Patsnap · data cut-off 2026-07-31 · downloaded from the live page
Patent Landscape · Recommender SystemsExploration and Bandit Methods Patents: Filing Trends and Leading Assignees
- 58.0% concentration. The top 5 of 47 ranked assignees hold 116 of 200 records in scope, with a long tail of single- and few-filing entrants behind them.
- Filing cooled from its 2019 peak. 2019 saw 34 records, the high point so far; 2021 to 2024 activity fell 18%, from 22 to 18 records, though 2025-2026 figures are still filling in under normal publication lag.
- AI computing dominates the IPC mix. G06N covers 62.0% of the 200 records in scope, well ahead of general data processing (G06F, 33.0%) and business-method filings (G06Q, 22.0%).
200
Published Records
58%
Top-5 Share of All Records
-18%
Filing Growth 2021→2024
US
Leading Jurisdiction
Filing growth compares 2021 (22 records) with 2024 (18) — 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 200 records in scope (CR5), not by the ranked leaders only.
Published byPatsnap Research··7 min readSourced from Patsnap Eureka
Overview
What this landscape covers
Exploration and bandit methods sit at the intersection of reinforcement learning and applied recommendation: contextual bandits, off-policy evaluation, propensity scoring and related variance-reduction techniques for estimating what a system would have earned under a policy it did not run. This landscape covers 200 published records filed or published between 2015 and the 2026-07-31 cut-off, drawn from assignees spanning cloud platforms, telecoms equipment vendors, streaming services and financial institutions.
The dataset skews toward computing-and-AI classifications rather than pure telecom or gaming applications, which tells you where claim density already sits and where it does not. Publication lag of roughly 18 months means the most recent one to two years understate real filing activity; treat any apparent recent-year drop with that in mind.
Filing activity and technology composition, 2015-2026
Source: Patsnap Eureka. Assignee ranking and totals. Derived from a Patsnap search on Exploration and Bandit Methods covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
Let an AI agent run this analysis on your own technology
Pick a task. Every answer cites the patents behind it.
10,000 free credits to start
The Data
Filing trends and technology composition
The trend line and the IPC breakdown below are drawn from the same 200-record set that anchors every figure on this page; together they show when activity peaked and which technical classes carry the claim volume today.
A 2019 peak followed by a real, if partial, cooldown
Filings rose from 4 records in 2017 to a peak of 34 in 2019, then eased; the 2021-to-2024 window shows an 18% decline, from 22 to 18 records. Because publication lags filing by about 18 months, 2025 and 2026 figures will keep rising as later publications catch up, so the apparent tail-off should not be read as the field slowing down.
AI-model computing leads, wireless and business-method uses trail
G06N (computing based on AI models) appears on 62.0% of the 200 records in scope, followed by G06F general data processing at 33.0% and G06Q business/commerce processing at 22.0%. Wireless (H04W, 15.0%) and digital transmission (H04L, 10.0%) show bandit and exploration techniques applied to network management, while recognition (G06K), video (H04N) and gaming (A63F) each sit under 5%, marking narrower but still active application niches.
Shares are the percentage of the 200 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Source: Patsnap Eureka. Filing trend and technology composition. Derived from a Patsnap search on Exploration and Bandit Methods covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.
Go deeper on Exploration and Bandit Methods with Eureka
This page is one run against one query. Ask Eureka your own question about exploration and bandit methods and every answer comes back with the patent numbers behind it.
Try EurekaKey Patents
Representative filing and most-cited records
Representative Filing
Training a reinforcement learning machine learning model
ROYAL BANK OF CANADA
One or more computer processors are used to train a reinforcement learning machine learning model, such as a contextual bandit machine learning model. A training dataset is inputted to the model, which is trained based on that dataset. During training, an entropy measure is determined, feedback is generated from it, and the model is further trained based on that feedback.Filed by Royal Bank of Canada, published 2025-11-20 as US20250356207A1.
Most-cited records in this landscape
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | WO2023091664A1 | Radio access network intelligent application manager | 237 |
| 2 | US20220124543A1 | Graph neural network and reinforcement learning techniques for connection management | 197 |
| 3 | US20170308609A1 | Multi-result ranking exploration | 143 |
| 4 | US20240259879A1 | Radio access network intelligent application manager | 122 |
| 5 | US20190272465A1 | Reward estimation via state prediction using expert demonstrations | 77 |
| 6 | US20180220061A1 | Real-time semantic-aware camera exposure control | 69 |
| 7 | US20200296741A1 | Virtual radio access network control | 55 |
| 8 | US20240273411A1 | Artificial intelligence regulatory mechanisms | 52 |
| 9 | US20200012681A1 | Personalizing explainable recommendations with bandits | 48 |
| 10 | US20240023028A1 | Wireless network energy saving with graph neural networks | 41 |
Citation counts favour older filings inside any searched corpus; read them as a signal of influence on later work, not as a measure of current commercial importance.
Each row carries its publication number; clicking a row searches Eureka by that number.
Source: Patsnap Eureka. Citation counts and representative records. Derived from a Patsnap search on Exploration and Bandit Methods covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
Run it yourself
Put your own technology through the same analysis
Where to run it
For builders
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 →Insights
What the numbers mean for filing strategy
Three patterns stand out once the ranking and the trend line are read together: concentration at the top, a cooldown that is at least partly a publication-lag artefact, and an IPC mix that says where claim space is already crowded.
Concentration
58.0% / top 5
of 200 records in scope
Five assignees hold more than half the field
The top 5 of 47 ranked assignees account for 116 of the 200 records in scope, a 58.0% share. The leader alone holds 34 records against a fifth-place figure of 17, so the gap between first and fifth is itself informative about how contested the space is at the very top.
Top 10 combined reach 74.0% of all 200 records.
Momentum
-18% (2021-2024)
filings, complete years only
A real cooldown from the 2019 peak, not a collapse
Filings peaked at 34 records in 2019 and eased to 18 by 2024, an 18% decline over the 2021-2024 window that can be treated as complete. Several of the largest historical filers show zero records in the latest year, but that pattern is consistent with publication lag as much as with reduced R&D investment.
2025-2026 counts will rise as later publications post.
Technology mix
62.0% G06N
of 200 records in scope
AI-model computing is the dominant class, not the only one
G06N carries almost two-thirds of records, but a third also carry G06F general data-processing classification and over a fifth carry G06Q business-method classification, meaning bandit and exploration claims are frequently paired with commerce or general computing claims rather than filed in isolation.
Classes overlap; shares sum to more than 100% of 200 records.
Eureka AI Agent
Looking for what nobody has claimed yet?
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to exploration and bandit methods, with the prior art for and against each one.
Source: Patsnap Eureka. Co-assignee relationships and derived observations. Derived from a Patsnap search on Exploration and Bandit Methods covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
Players
Who is filing, and where the field is still open
The ranked leaders span cloud and search platforms, telecom equipment vendors, and streaming and enterprise software firms; behind them sits a long tail of single- or few-filing entrants, which is where new claim space is most available.
Leader
34 records
top-ranked assignee
A single leader well ahead of the rest of the field
The top-ranked assignee holds 34 of the 200 records in scope, roughly double the fifth-place count of 17. That gap suggests a sustained, multi-year filing programme rather than a handful of opportunistic applications.
Based on the 47-company ranked list.
Mid-tier
17 at fifth
records held
A competitive second tier before the tail thins out
From fifth to tenth place the count drops from 17 to 5, meaning the field narrows quickly outside the top five. Firms in this band include telecom-equipment vendors and cloud platforms applying bandit methods to network and recommendation problems.
Top 10 combined hold 74.0% of all 200 records.
Long tail
37 of 47
assignees outside the top 10
Most ranked assignees hold a handful of records each
Thirty-seven of the 47 ranked assignees sit outside the top 10, each contributing a small number of the remaining 26.0% of records. This tail is where opportunistic, single-application filers and newer entrants show up.
Denominator is the 200 records in scope.
Recent-year filing momentum by assignee
| Assignee | Recent year | YoY |
|---|---|---|
| Microsoft Technology Licensing, LLC | 0 | -100% |
| DeepMind Technologies Limited | 0 | — |
| Alipay (Hangzhou) Information Technology Co., Ltd. | 0 | — |
| Telefonaktiebolaget LM Ericsson (publ) | 0 | -100% |
| Intel Corporation | 0 | — |
| Spotify | 0 | — |
| Huawei Technologies Co., Ltd. | 0 | — |
| BAST INC | 0 | -100% |
Source: Patsnap Eureka. Assignee-level momentum. Derived from a Patsnap search on Exploration and Bandit Methods covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
What's Next
Where to take this from here
The figures above describe the field as filed; turning that into a freedom-to-operate view or a filing plan means going claim by claim.
Map claims against your own architecture
Run your specific bandit, off-policy evaluation or propensity-scoring approach against the most-cited records above to see which claim elements are already occupied and which are open.
Explore claim-level analysis in EurekaTrack the leading assignees' recent filings
Several top-ranked assignees show zero records in the latest year; distinguishing a real slowdown from publication lag requires watching their filings over the next few quarters.
Set up assignee monitoring in EurekaDraft around the identified white space
The under-claimed sub-areas above are starting points, not conclusions; validate them against full claim text before committing R&D or filing budget.
Draft and compare claims in EurekaSource: Patsnap Eureka. Forward-looking reading of the same dataset. Derived from a Patsnap search on Exploration and Bandit Methods covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
FAQ
Common questions about this landscape
The ranking covers 47 assignees, with the leader holding 34 of the 200 records in scope and the top 5 combined holding 116 records, a 58.0% share of the field. The field narrows quickly after that: fifth place holds 17 records and tenth place holds 5, so the top 10 together account for 74.0% of all records. Beyond that, most of the remaining assignees hold only a handful of records each, forming a long tail of occasional filers rather than a second wave of sustained programmes.
Filings peaked at 34 records in 2019 and fell to 18 by 2024, an 18% decline over the 2021-2024 period that can be treated as a complete window. That is a genuine cooldown from the peak, but it should not be read as the field disappearing: publication lags filing by roughly 18 months, so 2025 and 2026 figures are still incomplete and will rise as later publications post. Several previously active filers show zero records in the latest year, which is consistent with that lag as much as with reduced investment.
G06N, computing based on AI models, covers 62.0% of the 200 records in scope and is the dominant classification. G06F general data processing (33.0%) and G06Q business and commerce processing (22.0%) are also common, often layered on top of a G06N classification on the same record. Smaller but active clusters sit in wireless networks (H04W, 15.0%), digital transmission (H04L, 10.0%), data recognition (G06K, 5.0%), video/pictorial communication (H04N, 4.5%) and games (A63F, 3.5%).
US20250356207A1, filed by Royal Bank of Canada and published 2025-11-20, covers training a reinforcement learning model, described in the filing as including a contextual bandit model, where an entropy measure computed during training generates feedback used for further training. Anyone building an entropy-based feedback loop for training a contextual bandit model in a similar sequence should review this filing's claim scope closely. It is a recent publication, so its practical blocking effect will become clearer as related continuations and office actions appear.
Application niches with lower claim density than the core computing and business-method cluster include video/TV recommendation (H04N, 4.5% of records), gaming and amusements (A63F, 3.5%) and data recognition pipelines (G06K, 5.0%). These are not empty, but they carry far fewer records than the dominant G06N and G06F clusters, suggesting more room for a narrowly drafted first claim. Any white-space read should be checked against full claim text rather than classification counts alone, since a thin IPC subclass can still sit behind a broad existing claim.
Answers are grounded in the same dataset. Derived from a Patsnap search on Exploration and Bandit Methods covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.Run this in Eureka MCP
Research Exploration and Bandit Methods in depth with Eureka
Go past this page: query the whole exploration and bandit methods 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.