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/hybrid-sparse-dense-retrieval-patent-landscape/ · Patsnap · data cut-off 2026-07-31 · downloaded from the live page
Patent Landscape · Vector Search & RAGHybrid Sparse Dense Retrieval Patents: Who Leads and Where the Claim Space Is Still Open
- 45.9% concentration. The top 5 assignees hold 273 of 595 records in scope — a field with a clear head and a long tail behind it.
- +322% filing growth. Filings rose from 9 in 2021 to 38 in 2024, the last year with complete publication data.
- 91.9% sit in G06F. Nearly all records carry a G06F digital-data-processing class, while AI-model (G06N) and speech (G10L) overlap remains comparatively thin.
595
Published Records
46%
Top-5 Share of All Records
+322%
Filing Growth 2021→2024
US
Leading Jurisdiction
Filing growth compares 2021 (9 records) with 2024 (38) — 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 595 records in scope (CR5), not by the ranked leaders only.
Published byPatsnap Research··7 min readSourced from Patsnap Eureka
Overview
What this landscape covers
Hybrid sparse-dense retrieval sits at the intersection of classical inverted-index search and learned dense embeddings, joined by score fusion techniques such as reciprocal rank fusion. The 595 records in scope were pulled using a search string that pairs core retrieval mechanisms — dense retrieval, inverted index, score fusion — against the practical problems that force hybrid designs: lexical matching gaps, query expansion, domain shift, index synchronization and term weighting. That pairing is deliberate: a patent that only claims a dense encoder or only claims an inverted index sits outside this set unless it also addresses one of the fusion or adaptation problems.
Filing activity spans 2015 through the middle of 2026, with the most recent one to two years understated because publication trails filing by roughly 18 months. Reading the trend as flat or declining in 2025–2026 would be a misreading of that lag, not a signal about the technology.
Filing activity, 2017–2026
Source: Patsnap Eureka. Assignee ranking and totals. Derived from a Patsnap search on Hybrid Sparse Dense Retrieval 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 Numbers
Filing trends and technology composition
Two views of the same 595 records: filing volume over time, and the IPC subclasses those records carry. Because a single record can be tagged with several classes, the class shares below sum to well over 100% — that is expected and is not a data error.
Filing trend, 2017–2026
Filings climbed from 29 in 2017 to a peak of 38 in 2024, including a +322% jump from 9 filings in 2021 to that 2024 peak. 2025 and 2026 figures are still filling in as publications catch up to filing dates.
Technology composition by IPC subclass
G06F covers 547 of 595 records (91.9%), confirming that nearly every filing in this set is framed as digital data processing at its core. G06N (AI models, 11.6%), H04W (wireless networks, 5.9%) and H04L (digital transmission, 5.4%) mark where retrieval intersects learned models and network delivery; G10L (speech, 2.4%) and G06V (image/video, 1.3%) show that multimodal hybrid retrieval is still a minority pursuit inside this corpus.
Shares are the percentage of the 595 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 Hybrid Sparse Dense Retrieval covering 2015–2026, data cut-off 2026-07-31. Counts reflect published records only and shift as new filings publish.
Go deeper on Hybrid Sparse Dense Retrieval with Eureka
This page is one run against one query. Ask Eureka your own question about hybrid sparse dense retrieval and every answer comes back with the patent numbers behind it.
Try EurekaKey Patents
Representative filing and most-cited prior art
Representative Record
US8620900B2 — Dual-index query expansion and relevance feedback
THE HONG KONG POLYTECHNIC UNIVERSITY
A method for using dual indices to support query expansion, relevance/non-relevance models, blind/relevance feedback and an intelligent search interface: an inverted index produces an initial ranked result list, a forward index is then used to identify the top N relevant information units, compute query expansion weights, and build relevance models from that feedback loop.Filed by The Hong Kong Polytechnic University, published 2013-12-31 — predates most of the dense-retrieval filings in this set and defines a dual-index pattern still referenced in hybrid designs.
Most-cited records in this landscape
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US6243713B1 | Multimedia document retrieval by application of multimedia queries to a unified index of multimedia data for … | 895 |
| 2 | US20080077570A1 | Full Text Query and Search Systems and Method of Use | 779 |
| 3 | US6847966B1 | Method and system for optimally searching a document database using a representative semantic space | 738 |
| 4 | US5920859A | Hypertext document retrieval system and method | 644 |
| 5 | US20160360382A1 | Systems and Methods for Proactively Identifying and Surfacing Relevant Content on a Touch-Sensitive Device | 590 |
| 6 | US20160360336A1 | Systems and Methods for Proactively Identifying and Surfacing Relevant Content on a Touch-Sensitive Device | 567 |
| 7 | US20100223276A1 | Automated Generation of Metadata for Mining Image and Text Data | 500 |
| 8 | US7983915B2 | Audio content search engine | 461 |
| 9 | US20060117002A1 | Method for search result clustering | 373 |
| 10 | US20080270138A1 | Audio content search engine | 365 |
Citation counts favour older filings simply by virtue of longer exposure; treat them as a measure of influence on later filers, not of current commercial relevance.
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 Hybrid Sparse Dense Retrieval 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 data means for filing strategy
Three patterns stand out once concentration, growth and class composition are read together.
Concentration
45.9%
of 595 records held by top 5
A short head, a long tail
The leader alone accounts for 72 records, and the top 5 combined hold 45.9% of all 595 records in scope. Below the top 10 (59.0% combined), the ranking thins into single- and double-digit filers, meaning most of the field's differentiation happens outside the leading names.
Read this as: dense competitive pressure at the top, open room further down the ranking.
Momentum
+322%
filing growth, 2021→2024
Growth concentrated in a three-year window
Filings rose from 9 in 2021 to 38 in 2024, the peak year in the dataset and the last year with complete publication coverage. That growth window coincides with the broader shift toward embedding-based retrieval, but the patent evidence only supports the 2021-2024 span — treat 2025 onward as incomplete rather than declining.
Any claim about a 2025/2026 slowdown is not supported by this data; it reflects the publication lag.
Class overlap
91.9%
of records in G06F
Core claims are digital-processing framed
With G06F present in 547 of 595 records, most applicants are still anchoring hybrid retrieval claims in general digital data processing rather than in AI-model-specific classes. G06N appears in only 11.6% of records, suggesting the AI-model framing of these inventions is a secondary classification rather than the primary claim structure for most filers.
Claims tied tightly to G06N-style model architecture remain comparatively rare in this set.
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 hybrid sparse dense retrieval, with the prior art for and against each one.
Source: Patsnap Eureka. Co-assignee relationships and derived observations. Derived from a Patsnap search on Hybrid Sparse Dense Retrieval 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 gate sits
The ranked list covers 100 companies counted by records, not a top-50 or top-100 cut of a larger universe — this is the full ranking the data endpoint returns for this search.
Leader
72 records
documents from the top assignee
A single filer well ahead of the pack
The leading assignee holds 72 of the 595 records in scope, more than double the fifth-place holder at 20. That gap defines the shape of the whole ranking: one dominant filer, a cluster of established technology and enterprise-search vendors behind it, then a long tail.
Recent-year momentum for several of the largest historical filers has slowed to zero in the latest year, per the momentum data.
Mid-field
20 records
fifth-place assignee
Enterprise search and legal-information vendors cluster mid-table
Assignees built around enterprise document search and legal/business information retrieval occupy several of the top-10 positions, reflecting the fact that lexical-to-dense hybridization was first commercially urgent in full-text legal and enterprise search before spreading to consumer search and assistants.
Co-assignee pairs in this segment recur, pointing to internal group filing rather than external partnerships.
Momentum
1 in latest year
most active filer, latest year
Latest-year activity has gone quiet across the board
Momentum data shows only one of the tracked leading assignees filing in the latest year, with the rest at zero and one showing a -100% year-on-year change. Given the roughly 18-month publication lag, this understates real filing activity rather than indicating the leaders have stopped working in the space.
Do not read flat latest-year counts as leaders exiting the field.
Recent-year filing momentum by assignee
| Assignee | Recent year | YoY |
|---|---|---|
| Apple Inc. | 1 | — |
| International Business Machines Corporation | 0 | — |
| Google LLC | 0 | — |
| Thomson Reuters Global Resources | 0 | — |
| Microsoft Technology Licensing, LLC | 0 | — |
| The Toronto-Dominion Bank | 0 | -100% |
| Thomson Reuters Enterprise Centre GmbH | 0 | -100% |
| Oracle International Corporation | 0 | -100% |
Source: Patsnap Eureka. Assignee-level momentum. Derived from a Patsnap search on Hybrid Sparse Dense Retrieval 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 analysis
The dataset points to specific next steps depending on whether the goal is freedom-to-operate, portfolio building or competitive monitoring.
Check freedom-to-operate against the dense head
With 45.9% of records held by five assignees, a freedom-to-operate review for any hybrid fusion mechanism should start with those five portfolios before broadening to the long tail.
Run a freedom-to-operate check in Eureka →Watch the under-claimed branches
Speech, image/video and edge-wireless overlaps remain thin relative to the G06F core, which is where a first-mover claim is more likely to stand without dense prior art in the way.
Explore white space in Eureka →Track momentum, not just headcount
Zero recent-year filings from several historically dominant assignees may reflect publication lag rather than retreat; a monitoring workflow that re-checks in 12-18 months will separate the two.
Set up assignee monitoring in Eureka →Source: Patsnap Eureka. Forward-looking reading of the same dataset. Derived from a Patsnap search on Hybrid Sparse Dense Retrieval 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
It refers to systems that combine classical sparse, lexical retrieval (inverted indices, term weighting) with learned dense-vector retrieval, joined by a fusion step such as reciprocal rank fusion. In this dataset, a filing only qualifies if it addresses both a core retrieval mechanism and a practical adaptation problem like domain shift, query expansion or index synchronization. That combination is what separates hybrid retrieval patents from patents that claim a dense encoder or an inverted index in isolation.
The ranking is led by a single assignee with 72 of the 595 records in scope, well ahead of the fifth-place holder at 20 records. The top 5 assignees combined hold 45.9% of all records, and the top 10 hold 59.0%, so the field has a clearly concentrated head. Behind that, the ranking extends across 100 companies, with most holding only a handful of records each.
Filings grew sharply from 9 in 2021 to 38 in 2024, a +322% increase over that span and the highest year recorded in the dataset. Figures for 2025 and 2026 appear lower, but that reflects the roughly 18-month lag between filing and publication, not an actual slowdown. Any conclusion about the field cooling off should wait until those later years have had time to fully publish.
G06F (electric digital data processing) appears in 91.9% of the 595 records, making it the near-universal primary classification for this technology. G06N (AI-based computing) appears in 11.6% of records, and H04W (wireless networks), H04L (digital transmission) and G06Q (business data processing) each cover a mid-single-digit share. Because records often carry multiple IPC codes, these percentages add up to more than 100%, which is expected.
Based on the IPC composition, branches like speech-and-audio integration (G10L, 2.4% of records), image and video hybrid retrieval (G06V, 1.3%) and wireless-edge dense retrieval (H04W overlap) carry comparatively little filing density relative to the G06F core. That does not guarantee an invention is patentable there, but it does mean less prior art sits directly on top of those specific combinations compared with mainstream text-based fusion claims.
Answers are grounded in the same dataset. Derived from a Patsnap search on Hybrid Sparse Dense Retrieval 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 Hybrid Sparse Dense Retrieval in depth with Eureka
Go past this page: query the whole hybrid sparse dense retrieval 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.