US 6775677

I'll search for authoritative information on this patent number, including any recent litigation activity.

I did not find any CAFC docket activity for this specific patent. Let me confirm with a couple of targeted searches.

US Patent 6,775,677 — Summary

Important disambiguation first: I checked specifically for litigation/CAFC-2026 activity on 6,775,677 and found none. Several search returns surface different patents with superficially similar numbers — do not conflate them. Notably:

• U.S. Patent No. 7,679,637 (Time-Shifted Web Conferencing) — the subject of the prominent Fed. Cir. Jan. 22, 2026 decision US Pat. No. 7,679,637 LLC v. Google LLC and related PTAB/mandamus proceedings. Unrelated.
• U.S. Patent No. 6,751,677 (Plectrum firewall/NAT patent), U.S. Patent No. 7,356,677 (GEMSA), the Hysitron '677 patent, and Apple's design patent D677 — all unrelated.

I cannot attest to the complete absence of activity in every 2026 docket, but targeted searches on the literal number produced nothing.

---

Bibliographic Data

Source: https://patents.google.com/patent/US6775677/en and https://uspto.report/patent/grant/6,775,677

---

Abstract (verbatim)

"To identify and describe one or more topics in one or more documents in a document set, a term set process creates a basic term set from the document set where the term set comprises one or more basic terms of one or more words in the document. A document vector process then creates a document vector for each document. The document vector has a document vector direction representing what the document is about. A topic vector process then creates one or more topic vectors from the document vectors. Each topic vector has a topic vector direction representing a topic in the document set. A topic term set process creates a topic term set for each topic vector that comprises one or more of the basic terms describing the topic represented by the topic vector. Each of the basic terms in the topic term set associated with the relevancy of the basic term. A topic-document relevance process creates a topic-document relevance for each topic vector and each document vector. The topic-document relevance representing the relevance of the document to the topic. A topic sentence set process creates a topic sentence set for each topic vector that comprises of one or more topic sentences that describe the topic represented by the topic vector. Each of the topic sentences is then associated with the relevance of the topic sentence to the topic represented by the topic vector."

---

Plain-Language Overview of the Independent Claims

All four independent claims (1, 14, 15, 16) recite essentially the same six-step pipeline, differing only in claim format (system / method / means-plus-function system / program product). Each also includes a specific limitation that the topic-document relevance be computed as the cosine between the topic vector and the document vector — a narrow "how" limitation that appears in every independent claim.

Claim 1 — System

A computer system (one or more CPUs and memories) implementing:

1. Term set process — build a "basic term set" of one- or multi-word terms from the document set.
2. Document vector process — create a document vector for each document; its direction represents what the document is about.
3. Topic vector process — derive one or more topic vectors from the document vectors; each topic vector's direction represents a topic in the set.
4. Topic term set process — for each topic vector, select the basic terms that describe that topic, each associated with a relevance value.
5. Topic-document relevance process — for each topic/document pair, compute relevance from the respective vector directions, specifically by computing the cosine between the topic vector and document vector.
6. Topic sentence set process — for each topic vector, produce a set of representative topic sentences, each with a relevance to the topic.

Claim 14 — Method

The same six steps framed as a method executing on a computer (term set → document vectors → topic vectors → topic term set → cosine-based topic-document relevance → topic sentence set). No hardware/means language.

Claim 15 — System (means-plus-function)

A system expressed as "means for creating…" corresponding element-by-element to the six steps of claim 1, including the cosine-based relevance means. Presumably intended to invoke 35 U.S.C. § 112(f) and capture the corresponding structures/processes disclosed in the specification.

Claim 16 — Computer Program Product

A program product that, when performing, carries out the same six steps (term set → document vectors → topic vectors → topic term set → cosine-based topic-document relevance → topic sentence set).

Note on scope: the dependent claims add the substantive details — claim 2 (term-document vectors, a conversion matrix, document/sentence vector conversion, and an interpretation matrix); claim 3 (topic terms from interpretation-matrix × topic-vector products); claim 4 (topic sentences by magnitude + direction similarity); claims 5–8 (the document map visualization, hover-based color syncing across maps, direction indicator, click-to-open full text); claims 9–13 (topic sentence frame and its interactive highlighting). Notably, the distinctive conversion-matrix / interpretation-matrix encoding and the document-map visualization described in the specification appear only in dependent claims, not in the independent claims.

---

Litigation posture

• Assignee: IBM, original assignee of record (assignment recorded; inventors Ando, Boguraev, Byrd, Cooper, Neff).
• USPTO maintenance: fees paid at years 4, 8, and 12; fee recorded Feb. 10, 2016 (12-year payment) — final payment; patent expired March 2, 2020 (patents filed before the 1995 GATT-URAA transition can run 20 years from filing, i.e., to 2020-03-02).
• Litigation / CAFC 2026 dockets: I found no records tying patent 6,775,677 to any 2026 Federal Circuit docket, PTAB proceeding, or district-court action. Given the patent is expired, active enforcement in 2026 is unlikely. Uncertainty: negative findings from web search are not proof of absence; a definitive check would require a full Docket Navigator / PACER / PTAB API query, which I cannot run here.

If you want, I can attempt narrower queries (e.g., a specific district or party name) to try to confirm the "no litigation" conclusion.

I'll research litigation involving US patent 6775677. Let me search multiple sources.

Let me search more targeted litigation sources for this specific patent.

Let me check litigation-specific databases and the Unified Patents portal directly.

Litigation Search Results for U.S. Patent No. 6,775,677

Bottom line: I found no evidence of any litigation (district court, ITC, or PTAB) involving U.S. Patent No. 6,775,677.

What I searched

I searched for the exact number "6775677" / "6,775,677" in combination with litigation-related terms, plus checks against litigation databases and the Unified Patents portal. Across these queries, the patent surfaced only in prosecution/bibliographic contexts, never in an enforcement or litigation context:

• Google Patents — the patent's full text, claims, citations, and "Cited By" lists (54 examiner-cited + 91 third-party citations).
• uspto.report/patent/grant/6,775,677 — bibliographic/prosecution record only.
• Unified Patents litigation portal — no responsive results for this number.

Notably, the patent's Google Patents record shows no "Litigation" section, and its legal-status timeline shows only ordinary prosecution and maintenance-fee events (granted 2004-08-10; maintenance fees paid; anticipated expiration 2020-03-02; status "Expired - Lifetime"). There is no indication it was ever asserted.

⚠️ Do-not-conflate warnings (similar-number false positives)

My searches returned several cases involving different patent numbers that must not be attributed to '677:

• U.S. Patent No. 6,842,677 — Vehicle Interface Technologies, LLC v. Jaguar Land Rover North America, LLC, C.A. No. 14-339-RGA (D. Del., filed 2014). This is a different patent ('677 vs. '677 — literally 6,842,677, not 6,775,677). Not relevant.
• U.S. Patent No. 7,356,677 — asserted by GEMSA against multiple AWS users (E.D. Tex., 2016, cause nos. 2:16-cv-0023 through -0037). Different patent. Not relevant.
• U.S. Patent No. 5,677,537 — cited as prior art in an unrelated PCT search report. Not relevant.
• Various state/financial records (e.g., a North Carolina securities register listing "6775677 Gonzalez Alejandro Jr") — not patent litigation.

Per your standing instruction, I am interpreting identifiers literally and not auto-correcting them; none of the above are U.S. Patent No. 6,775,677.

Caveats and confidence

• Confidence: Moderately high that no litigation exists, but this is not exhaustive. My searches hit a step limit before I could query CourtListener/PACER and Docket Navigator directly. Free/aggregator sources (Google Patents, Unified Patents, general web) are strong indicators but are not a substitute for a full commercial docket search.
• If the patent was ever asserted, it would most plausibly have occurred 2004–2020 (pre-expiration). IBM, as assignee, historically enforced some patents, but I found no record tying this specific patent to any suit.
• If you need a legally conclusive answer, I'd recommend running the exact string "6,775,677" through a paid docket database (Docket Navigator, Lex Machina) and PACER/CAFC dockets directly, as those were not fully queried here.

Recommendation: State for the record that, based on available search results, no litigation involving U.S. Patent No. 6,775,677 is known, while flagging the limited scope of the search.

I'll verify the structured data with targeted searches before finalizing.

Let me run a couple more targeted checks to be thorough.

PTAB / AIA Trial Proceedings — US Patent 6,775,677

Verification note: I ran the canonical source first — the USPTO Open Data Portal "PTAB proceedings on file" block for this patent — and it returns zero AIA trial proceedings. I then ran independent web searches against PTAB E2E/PTACTS filings, Federal Circuit dockets, and CourtListener. Every "677" hit I surfaced during searching belongs to a different patent — e.g., Fractus's antenna patent 11,031,677 (IPR2025-01026, IPR2026-00191), Wintek's 8,294,677 (IPR2013-00433), and a messaging patent 8,874,677. None of those are US 6,775,677, and I have not carried them into this report. Do not let a docket-search vendor conflate the two.

---

Proceedings overview

Total AIA trial proceedings on US 6,775,677: 0 — 0 active, 0 claims invalidated, 0 claims sustained, 0 settled, 0 institution denials. The patent has never been challenged at the PTAB, in any forum type (IPR, PGR, or CBM), at any point in its ~20-year enforceability life. The defensive posture this gives a defendant is unusual and should be read carefully: it does not mean the patent is "hardened" by surviving attacks — it means every one of claims 1–16 is completely untested before the Board. There is no FWD to lean on, no claim-cancellation certificate to point to, and no petitioner estoppel on the docket. The one external data point that matters more than the empty PTAB list is the patent's legal status: Expired – Lifetime, anticipated expiration 2020-03-02. A defendant today is facing a time-barred forward-looking case — the only live exposure is a § 286 six-year back-damages window for past infringement, and even that is shrinking.

---

Per-proceeding detail

None exists. Because the instructions call for a per-proceeding format, I'll state the finding for the record rather than skip it:

(no proceedings) — No petitioner v. International Business Machines Corp.

• Type: N/A — no IPR, no PGR, no CBM ever filed against US 6,775,677.
• Filed: N/A
• Status: No AIA trial activity on file (USPTO ODP structured data); independently corroborated by web search across PTAB filings and Federal Circuit/CourtListener dockets.
• Judge panel: N/A — no panel has ever been assigned.
• Petition grounds: N/A
• Institution decision: N/A
• Final Written Decision: N/A — no claim of this patent has ever been construed, invalidated, or confirmed by the Board.
• Settlement / termination: N/A
• Appeal: N/A — no FWD to appeal. I found no CAFC appeal involving this patent number.
• Defensive value: You would be the first challenger. That is a double-edged sword: no adverse claim-construction record exists to exploit, but also no free kill. Your validity attack starts from scratch — a full prior-art search and IPR budget from zero.

---

Strategic summary

Claim status: everything is UNTESTED. Claims 1–16 as issued (including the certificate of correction dated 2005-09-20) remain presumptively valid and unadjudicated — not because they survived a PTAB challenge, but because no PTAB challenge was ever filed. The independent claims are claim 1 (system: term set → document vector → topic vector → topic term set → topic-document relevance → topic sentence set), claim 14 (method counterpart), claim 15 (means-plus-function system counterpart), and claim 16 (computer program product counterpart). The dependent claims 2–13 add document/sentence vectorization via a conversion matrix and interpretation matrix (2–4) and the "document map" graphical UI features (5–13). There is no certificate of cancellation, no amended or substitute claim, and no narrowed claim scope on the record. Any assertion built on any of these claims is built on unreviewed terrain.

Estoppel landscape: empty. § 315(e)(2) estoppel attaches only to a petitioner (and its privies/RPIs) who was party to an instituted IPR that reached a final written decision. With no IPR ever instituted, there is no IPR estoppel against anyone — including you. Every prior-art ground is available: § 102(a)/(b)/(e) and § 103(a) under pre-AIA law (this patent has a 2000-03-02 priority date, so pre-AIA §§ 102/103 and pre-AIA § 112 ¶ 6 govern the means-plus-function claims 15 and the "mechanism" language in dependent claims). Nothing is blocked. There is also no In re Vivint / SAS / Fintiv procedural precedent specific to this patent to navigate.

Pattern signals. There is no petitioner to profile and no patent-owner appeal history, because the docket is empty. There is no defensive aggregator (Unified Patents, RPX, etc.) in any chain — a notable absence, since Unified's usual pattern is to file on precisely this flavor of broad, expired-in-a-decade software patent. Two structural facts explain the silence and should shape your strategy: (1) the patent expired 2020-03-02, so after that date there is no ongoing infringement to enjoin and no reason for a competitor to spend ~$500K–$800K on an IPR just to stop an injunction; and (2) IBM's assertion campaigns (e.g., IBM v. Groupon) targeted other patents in its portfolio, not this one. The absence of PTAB activity is best read as "nobody bothered" rather than "this patent is a fortress."

---

Recommended next steps

• If you are a defendant being asserted today: The patent expired 2020-03-02. Confirm the priority/term math yourself against the § 154 term and any PTA, because that expiry date is your single most valuable fact. An IPR is only worth filing if the case is economically live — institution on an expired patent is routinely disfavored on discretionary grounds (settled expectations, no forward-looking need, finite Board resources), and the Board has been denying such petitions. Focus instead on § 286 damages limitations and the invalidity fight in the district court, where you face no estoppel because there is no IPR record.
• Because there is no FWD, do not cite one. If opposing counsel or a search vendor hands you an IPR number purporting to cover this patent, verify the patent number digit-for-digit — the '677 suffix is heavily shared (Fractus 11,031,677; Wintek 8,294,677; messaging 8,874,677). Google Patents confirms the correct record at https://patents.google.com/patent/[US6775677B1](/patent/US6775677B1)/en (16 claims).
• If you want to build a validity record: Commission a fresh pre-AIA § 102/§ 103 search against the 2000-03-02 priority date. The patent's core concepts — SVD/LSI document vectors, cosine topic-document relevance, eigenvector-based topic detection, and eigenvector-cosine sentence selection — were dense in the 1996–1999 IR literature (Latent Semantic Indexing, Salton's vector space model, the Ding SIGIR '99 reference the patent itself incorporates). That literature is a natural starting point and is not blocked by any estoppel.
• Primary sources to monitor: USPTO PTAB E2E / PTACTS (https://ptacts.uspto.gov/ptabweb/) for any newly-filed petition, and CourtListener (https://www.courtlistener.com) for CAFC appeals. Given the 2020 expiry, I would rate new IPR activity as improbable — but if the patent owner monetizes the back-damages tail, a defendant could still file, so re-check before relying on this report.

Bottom line: There is no PTAB activity on file for US 6,775,677, and I found no unreported proceeding in web searches. The patent is expired, unadjudicated, and free of IPR estoppel for any challenger — which cuts against the patent owner, not for them, if they come after you now.

I'll verify this patent's ownership record against the authoritative sources before writing anything up.

Let me check the dedicated patent-ownership trackers and confirm whether any post-issuance assignment exists.

I was unable to load assignmentcenter.uspto.gov directly in this session — the queries returned no assignment records for this patent number. What follows is grounded in the Google Patents legal-events feed (which mirrors the USPTO Assignment Center underlying data) for US 6775677, plus targeted searches. Where a field is not in the record, I say so rather than guess.

Verification links: https://assignmentcenter.uspto.gov/ · https://patents.google.com/patent/[US6775677B1](/patent/US6775677B1)/en

Inventors

All five named inventors are listed on the single recorded assignment as assignors running to International Business Machines Corporation. The recorded signing dates span 2000-01-21 to 2000-03-02 (per the assignment free-format text: SIGNING DATES FROM 20000121 TO 20000302), i.e., the inventors executed within roughly six weeks of the 2000-03-02 filing date. This is the ordinary inventor-to-corporate-employer pattern and shows no unusual departure/attrition signal — I found no evidence any inventor left IBM within 12 months of filing, and none of the five appears in the assignment record again in any capacity.

I could not independently confirm each inventor's specific IBM lab (e.g., T.J. Watson Research Center) from the assignment record alone; the record establishes employer = IBM, not the site.

Original assignee

International Business Machines Corporation (Armonk, New York), a New York corporation — the assignee named on the issued patent and the only recorded owner.

• Primary line of business: enterprise computing — hardware, software, and IT services, with a large industrial research arm.
• Product embodying the claims: not determinable from the ownership record. The patent claims a method/system for detecting and describing topics across a collection of documents (topic vectors, interpretation matrix, topic term sets, topic sentence sets, cosine-based topic-document relevance — claims 1 and 14). This reads on text-mining/search and analytic functionality IBM Research worked on in the 2000 timeframe, but the assignment and legal-events records contain no product, licensing, or commercialization data, so I will not assert a shipping product.
• Current status: operating. IBM has not been dissolved, acquired, or in bankruptcy. The patent itself is Expired – Lifetime, with anticipated expiration 2020-03-02 (20 years from filing). Maintenance fees were paid through the year-12 payment recorded 2016-02-10, consistent with the patent running to its natural term.

Assignment timeline

Exactly one assignment is recorded against this patent. There are no post-issuance transfers of any kind — no shell-entity assignment, no security interest, no merger, no name change, no license recordation.

• 2000-01-21 to 2000-03-02 (executed) / recorded 2000-03-02 — Reel 010665/0155
  • Conveyance: Assignment (assignment of assignors' interest)
  • Assignor: Rie Kubota Ando; Branimir K. Boguraev; Roy Jefferson Byrd; James William Cooper; Mary Susan Neff
  • Assignee: International Business Machines Corporation, New York
  • Correspondent: not disclosed in the record I could retrieve. IBM filed this through its own in-house assignment practice; the reel/frame text does not surface an outside recording correspondent, and I will not guess one.
  • Context: original inventor-to-employer assignment at filing — not an acquisition, fire-sale, reorg, or transfer-to-asserter.

Post-2000 legal events are maintenance/administrative only (fee-payment procedure 2003-12-08; patent grant 2004-07-22; certificate of correction 2005-09-20; maintenance fees 2007-09-19, 2012-06-29, 2016-02-10; expiration 2020-03-02). None of these is a change of ownership.

Note on a search artifact: web queries surfaced IBM→Lenovo assignment records at reels 3080/0149, 3805/0456, 3148/0353, and 5684/0088. Those are IBM trademark / IBM Asia Products–to–Lenovo records in TTAB abstract-of-title documents — they are not assignments of US 6775677 and must not be read into this chain.

Timeline diagram

timeline
    title Ownership of US 6775677
    2000 : Filed by five IBM inventors
         : Assigned to IBM on reel 010665/0155
    2004 : Patent issued as US 6775677 B1
    2016 : Final maintenance fee paid
    2020 : Expired at end of 20 year term

NPE / troll-pattern signals

1. Shell-entity transfer — Not present. No assignment to any "IP / Patents / Licensing / Holdings / Ventures" entity exists. The only recorded assignee is IBM (reel 010665/0155), and no subsequent reel/frame appears in the chain.

2. Known asserter in the chain — Not present. IBM is an operating company, not on the Acacia / Marathon / IV / Wi-LAN / Conversant / Vringo / Pendrell / Round Rock / MPHJ lists. I also checked the Stanford NPE Litigation Database: patent 6775677 returns no cases (the only near-match returned was 6,777,757, a different patent). No assignee in this chain matches a public NPE list.

3. Repeat correspondent across the chain — Not present / unverifiable. With only one recorded link and no correspondent named in the retrievable record, there is no recurrence to cite. Absence of a correspondent is not evidence of an NPE pattern.

4. Cascading transfers — Not present. Zero transfers in 20 years, versus the pattern's "<24 months through chained LLCs."

5. Pre-litigation transfer — Not present. No assignment within 6 months of any infringement suit; no suit naming this patent was found, and no transfer occurred after the original 2000-03-02 record.

6. Bankruptcy fire-sale — Not present. IBM has never filed Chapter 7/11, and no bankruptcy-sale conveyance is recorded.

7. Privateering — Not present. No operating-company-to-NPE transfer; IBM retained the patent to expiry.

8. Defensive aggregator (anti-NPE) — Not present as such. The chain does not terminate at RPX, AST, LOT, Unified Patents, or OIN. The patent was effectively neutralized instead by term expiration on 2020-03-02, which is a different mechanism.

Verdict

Insufficient data — but with a benign explanation, not an evidentiary gap: the record contains only the original assignment (reel 010665/0155, recorded 2000-03-02, inventors → IBM). Per the rubric, "only the original assignment" maps to Insufficient data, but the substance is that this is an operating-company-owned patent with no NPE activity in its chain. IBM held US 6775677 continuously from its 2000-03-02 filing through natural expiration on 2020-03-02, paying maintenance fees through 2016, with no recorded post-issuance assignment, no shell-entity transfer, no asserter in the chain, and no litigation surfaced in the NPE databases checked. There is no NPE or troll-pattern finding here.

Caveat: I could not directly query the live USPTO Assignment Center in this session, and the correspondent-of-record field for reel 010665/0155 was not retrievable. If you need the recorded correspondent and the on-file assignment image (which would show the executing IBM attorney/signatory), pull reel 010665/0155 directly at https://assignmentcenter.uspto.gov/ — but nothing in the existing record suggests that image will reveal a second link, because no second reel/frame exists for this patent.

I'll research the cited references for US 6,775,677 to build a proper prior-art analysis. Let me run several targeted searches.

Let me gather the remaining cited references and the non-patent literature.

Prior-Art Analysis — US 6,775,677 B1

A. Target confirmation and search method

Target identified: US 6,775,677 B1 — "System, method, and program product for identifying and describing topics in a collection of electronic documents," application 09/517,540, filed 2000‑03‑02, granted 2004‑08‑10, assignee IBM. Front-page prior-art keywords: topic, document, vector, sentence, term.

Per your instruction I searched the literal number 6775677 only. I did not return anything for look-alike numbers (e.g., 7,679,637; 6,751,677; 7,356,677; D677), consistent with the disambiguation already established in the prior sections.

Method note / transparency: I do not have a live USPTO PatentCenter/Open-Data API in this session. I worked from (i) the authoritative full text of the '677 grant supplied in the prompt (the Google Patents mirror of the USPTO grant), and (ii) USPTO-sourced mirrors for the cited references — patents.google.com, uspto.report, patentimages.storage.googleapis.com. I could not run an exhaustive assignment/docket query.

The references below are exactly the six U.S. patent citations and the one non‑patent citation listed on the face of US 6,775,677 (the artwork the examiner considered).

---

B. Legal framework (dates matter)

US 6,775,677 was filed 2000‑03‑02 and was never subject to the AIA first‑inventor‑to‑file regime, so pre‑AIA 35 U.S.C. § 102 governs. With a filing date of 2000‑03‑02, the one-year bar date is 1999‑03‑02:

---

C. Reference-by-reference analysis

C‑1. US 5,857,179 A — Vaithyanathan, Adler, Hill (Digital Equipment Corp.)

"Computer method and apparatus for clustering documents and automatic generation of cluster keywords"

• Full citation: U.S. Patent No. 5,857,179, appl. 08/709,755, filed 1996‑09‑09, granted 1999‑01‑05, assignee Digital Equipment Corporation.
• § 102 category: § 102(b) (granted before 1999‑03‑02).
• Disclosure (brief): Builds an initial document‑by‑term matrix, each document an M‑dimensional vector; reduces dimensionality (PCA / SVD, selecting "k" loading vectors and projecting the initial matrix onto them); clusters the reduced vectors (k‑means); and automatically generates cluster keywords by computing, for each resultant vector, sums of products of vector components and Eigenvector components, then selecting the terms with the largest impact as "cluster summaries."
• Potential § 102 mapping:
  • Claim 2 — close on the term‑document vector → conversion matrix (SVD/PCA) → reduced document vector steps, but claim 2 also requires a term‑sentence vector process, a sentence vector process, and an interpretation matrix; Vaithyanathan discloses none of the sentence‑level machinery. → Not a full anticipation of claim 2.
  • Claim 3 — the "terms having greatest impact" computation (products of reduced‑vector components with Eigenvector components) is the nearest prior art to claim 3's "products of the interpretation matrix and the topic vector." But claim 3 depends from claim 2 (which requires the interpretation matrix), so it is not anticipated as claimed; it is the strongest § 103-type reference against claims 2 and 3.
  • Claims 1/14/15/16 — does not disclose deriving topic vectors from document vectors such that every document is close to a topic vector, cosine topic‑document relevance, or a topic sentence set. → No anticipation.
• URLs: https://patents.google.com/patent/[US5857179A](/patent/US5857179A)/en ; https://uspto.report/patent/grant/5,857,179

---

C‑2. US 5,794,178 A — Caid & Carleton (HNC Software, Inc.)

"Visualization of information using graphical representations of context vector based relationships and attributes"

• Full citation: U.S. Patent No. 5,794,178, appl. 08/632,519, filed 1996‑04‑12, granted 1998‑08‑11; divisional of 08/561,167 (now 5,619,709), in turn a continuation of 08/124,098 filed 1993‑09‑20. Assignee HNC Software, Inc.
• § 102 category: § 102(b) (granted before 1999‑03‑02).
• Disclosure (brief): Generates context vectors for words, documents and queries; forms a normalized summary vector per document (≈ the '677 "document vector"); measures similarity by dot product of vectors; reduces dimension by SVD/PCA; and visualizes documents as icons whose position is set by the dot product of the document summary vector with a coordinate context vector and whose attributes (size, color, distortion, etc.) encode further relevance. Clicking/double‑clicking an icon opens the associated document text.
• Potential § 102 mapping:
  • Claims 5–8 (document map / visualization) — this is the most relevant reference: icon position from a dot‑product relevance score, icon color/attribute from relevance, and click‑to‑open full text. It does not show the specific topic‑vector framing (its axes are user‑selected "coordinate words/query terms," not topic vectors derived from the document set), nor cross‑map hover color‑syncing (claim 6) or a direction arrow (claim 7). → Not a full anticipation of claims 5–8, but the closest art.
  • Claim 2 — SVD/PCA reduction of document vectors is disclosed; sentence vectors and interpretation matrix are not. → No anticipation.
  • Claims 1/14/15/16 — no topic‑vector process, no topic term set, no topic sentence set. → No anticipation.
• URLs: https://patents.google.com/patent/US5794178 ; http://www.everypatent.com/comp/pat5794178.html

---

C‑3. US 5,774,888 A — Light (Intel Corporation)

"Method for characterizing a document set using evaluation surrogates"

• Full citation: U.S. Patent No. 5,774,888, filed 1996‑12‑30, granted 1998‑06‑30, assignee Intel Corporation. (The '677 specification refers to this in its Background as the Light "evaluation surrogates" patent, which "needs the topic profiles predefined.")
• § 102 category: § 102(b) (granted before 1999‑03‑02).
• Disclosure (brief): Builds a compact token/term representation of each document (stopping/stemming deferred to preserve compound terms), defines pre‑defined "topic profiles," computes a similarity function between each document representation and each topic profile to yield a relevance measure, collects the measures in an evaluation surrogate, and plots documents as points in a "topic space" for visual analysis.
• Potential § 102 mapping:
  • Claims 1/14/15/16 — the topics are externally predefined (topic profiles), not topic vectors derived from document vectors, and there is no cosine topic‑document relevance, no topic term set, and no topic sentence set. → No anticipation. Indeed the '677 Background expressly distinguishes this reference.
  • Claims 5–8 — the topic‑space point plot is a distant relative of the document map, but the positioning/coloring scheme differs. → Not a full anticipation.
• URLs: https://patents.google.com/patent/US5774888#1 ; http://www.everypatent.com/comp/pat5774888.html

---

C‑4. US 5,937,422 A — Nelson, Schone, Bates (U.S. National Security Agency)

"Automatically generating a topic description for text and searching and sorting text by topic using the same"

• Full citation: U.S. Patent No. 5,937,422, appl. 08/834,263, filed 1997‑04‑15, granted 1999‑08‑10, applicant the United States (NSA).
• § 102 category: granted 1999‑08‑10 — less than one year before the 2000‑03‑02 filing, so not § 102(b); it is § 102(a) (patented before the invention) and, as a U.S. patent on an application filed 1997‑04‑15, also § 102(e) art effective 1997‑04‑15.
• Disclosure (brief): Stems input words; scores each word (part‑of‑speech, "language salience," frequency); expands each word through dictionary definition tree structures; collapses trees to word lists; combines into a final list; and chooses the top‑N scoring words as the "topic description." Documents can be searched/sorted by comparing the similarity of their topic descriptions.
• Potential § 102 mapping:
  • Claim 1(e)/claim 3 (topic term set) — this is prior art to the idea of automatically producing a set of topic‑defining terms; but it uses dictionary expansion and static word scores, not "products of the interpretation matrix and the topic vector," and it has no document/topic vectors at all. → No anticipation of claim 3 (which depends on the vector machinery of claim 2).
  • Claims 1/14/15/16 — no document vectors, no topic vectors, no cosine topic‑document relevance, no topic sentence set. → No anticipation.
• URLs: https://patents.google.com/patent/[US5937422A](/patent/US5937422A)#1 ; https://cryptome.org/nsa-vox-pat.htm

---

C‑5. US 5,696,533 A — Hewlett-Packard Company

"Method for selecting an item on a graphics screen"

• Full citation: U.S. Patent No. 5,696,533, filed 1995‑04‑05, granted 1997‑12‑09, assignee Hewlett‑Packard Company (related family: US 5,555,003; JP 3481296 B2).
• § 102 category: § 102(b) (granted before 1999‑03‑02).
• Disclosure (brief): Computer‑graphics picking/selection: an "item buffer" stores a unique item identifier per pixel; a pointer device selects/rolls over an object; the identified object is highlighted, and the operation is repeated until a selection is made. This is a graphics‑interface selection/highlighting disclosure, unrelated to text analysis.
• Potential § 102 mapping:
  • Relevant only to the interaction‑mechanics dependent claims — claims 5–8 (pointer hover/positioning + click) and claims 11–13 (highlighting of an item on pointer action). Even there it is a generic graphics‑selection teaching, not the document/topic‑specific structures recited. → No anticipation of any claim; possible § 103 support for the UI touch‑and‑feel limitations.
• URLs: https://patents.google.com/patent/US5696533#1 ; https://www.freepatentsonline.com/[5555003](/patent/5555003).html (family)

---

C‑6. US 6,466,240 B1 — Vadim Maslov

"Method for visually writing programs or scripts that transform structured text presented as a tree"

• Full citation: U.S. Patent No. 6,466,240 B1, filed 1998‑07‑08, granted 2002‑10‑15.
• § 102 category: granted after the '677 filing, so not § 102(a)/(b). It is potentially § 102(e) art only, effective as of its filing date 1998‑07‑08.
• Disclosure (brief): A visual/graphical method for authoring programs or scripts that transform structured text represented as a tree. There is no topic identification, no document/topic vector model, no relevance computation, and no summarization.
• Potential § 102 mapping: None. No claim of US 6,775,677 is anticipated; its subject matter is remote. It appears to have been cited as generic "graphical manipulation of text displayed as a tree" art.
• Note: Its presence on the face suggests only a peripheral (interface/representation) relevance.

---

C‑7. Non‑patent literature — Chris H. Q. Ding

• Full citation: Chris H. Q. Ding, "A Similarity‑based Probability Model for Latent Semantic Indexing," Proceedings of SIGIR '99, 22nd International ACM Conference on Research and Development in Information Retrieval, pp. 58–65 (1999). Listed on the face as the sole Non‑Patent Citation.
• Date / § 102 category: published 1999 (conference proceedings), before the 2000‑03‑02 filing but within one year, so § 102(a) (printed publication), not § 102(b).
• Disclosure (brief): A probabilistic (Gaussian‑log‑likelihood) model for Latent Semantic Indexing — i.e., a statistical objective for choosing the reduced dimension / weighting in a latent‑semantic (SVD) factorization of a term‑document matrix.
• Potential § 102 mapping:
  • The '677 specification expressly invokes this Ding paper as an example of the objective function g(.) used to choose k and of f(.) used to choose the scaling factor q in the conversion‑matrix process (claim 2 territory). It is therefore art to claim 2's conversion‑matrix/truncation step only, and only as a computational detail. It discloses nothing about topic vectors, topic term sets (claim 3), topic‑document relevance by cosine, or topic sentence sets. → No anticipation of any claim; § 103 material against the dimensionality‑reduction limitation of claim 2.
• URL: https://patents.google.com/patent/US6775677/en (reference listed under "Non‑Patent Citations")

---

D. Summary ranking (most → least relevant)

---

E. Bottom line

• No cited reference anticipates any claim of US 6,775,677 under § 102. The independent claims (1, 14, 15, 16) each require the combination of (a) a topic‑vector process deriving topic vectors from document vectors so that every document is close to a topic, (b) topic‑document relevance computed specifically as a cosine between topic and document vectors, (c) a topic term set, and (d) a topic sentence set — no single reference supplies all four.
• The most relevant prior art is US 5,857,179 (document vectors + SVD/PCA reduction + automatic keyword/topic‑term generation) and, for visualization, US 5,794,178 (dot‑product‑based icon positioning/coloring and click‑to‑open text). Together they are the natural § 103 challenge to claims 2, 3, and 5–8, which are where the '677 specification's distinctive conversion‑matrix/interpretation‑matrix encoding and document‑map visualization actually live (recall from the prior section that those features are not in the independent claims).
• US 5,937,422 and the Ding paper are narrower art: "topic description by ranked terms," and "LSI truncation objective," respectively — relevant to claim 3 and claim 2's k/q selection but not anticipatory.

---

F. Flags & caveats

1. Nothing here is legal advice, and my § 102 mapping is an analyst's first pass, not an invalidity opinion.
2. Identifier literalness (as instructed): the '677 text renders the Light/Intel "evaluation surrogates" reference inconsistently in at least one spot — the front‑page citation table lists US 5,774,888, while the Background narrative refers to the Light reference by number in a way that appears to differ by digits from the front‑page number. I have reported the front‑page number (US 5,774,888, confirmed against uspto.report/patent/grant/5774888) as controlling and not auto‑corrected either rendering. Recommend verifying against the original grant PDF.
3. Date sourcing: Where the prompt's front‑page citation table gave dates, I used them verbatim (e.g., US 5,696,533 — filed 1995‑04‑05, issued 1997‑12‑09; US 5,794,178 — priority 1993‑09‑20, issued 1998‑08‑11). I did not independently re‑verify every date against an official USPTO record in this session.
4. "Anticipates" vs. "relevant to": In strict § 102 terms, single‑reference anticipation fails for every cited reference. I have therefore also flagged each reference's § 103 (obviousness) significance, which is the more realistic role for this art.
5. Not re‑run here: a definitive prior‑art sweep would also include non‑cited art (e.g., Deerwester et al.'s LSI; Salton's vector‑space model; Cutting et al. Scatter/Gather, which appears among the "Similar Documents" on the '677 record) — if you want, I can extend the analysis to those and to the third‑party "Cited By" family.

If you'd like, I can (a) produce a claim‑chart (element‑by‑element) for the two strongest references against claims 1/2/3/5–8, or (b) attempt narrower USPTO queries by application number (09/517,540) or by the examiner's citation record to confirm the reference set.

I'll ground this § 103 analysis in the references actually cited on the patent's face, and verify their disclosures before mapping them to claim elements.

Obviousness Analysis — US 6,775,677 B1 under 35 U.S.C. § 103

Analyst's note on scope and honesty. This is a hypothetical/retrospective § 103 analysis. The patent expired March 2, 2020 (confirmed in the prior section), so there is no live invalidity defense value — but the analysis below is evaluated as of the March 2, 2000 filing date. I grounded the references using the "Cited By / Citations / Non-Patent Citations" material on the patent page plus the retrieved full text/abstracts of the cited references. Where I map a claim element to a reference based on an abstract or summary rather than the reference's full specification, I flag it. I do not assert that any single reference alone anticipates claim 1; the question asked is obviousness by combination.

---

1. The governing framework

• Pre-AIA § 103 applies. The application was filed March 2, 2000, before the AIA first-to-file date of March 16, 2013. So the pertinent references must qualify under pre-AIA § 102(a)/(b)/(e).
• Graham v. John Deere factors: (1) scope/content of prior art; (2) differences between prior art and claims; (3) PHOSITA level; (4) secondary considerations.
• KSR Int'l v. Teleflex (2007) governs the "motivation to combine" inquiry: a combination is obvious when the prior-art elements work "according to their established functions" and the combination is "the predictable use of prior art elements," with articulated reasoning (design incentive, market force, known technique to improve similar devices).
• Specification admissions count as prior art. Critically, the '677 specification itself admits the state of the art: clustering programs with cluster labels, hit lists, context-vector visualization, and "existing text summarization methods... generally based on sentence extraction or sentence generation." These admissions (and the "Problems with the Prior Art" section) do much of the work of supplying the problem and motivation.

---

2. Level of ordinary skill (PHOSITA)

As of March 2000: a person with an M.S. or Ph.D. (or equivalent experience) in computer science / information retrieval / computational linguistics, familiar with (i) the vector-space model of IR, (ii) Latent Semantic Indexing and SVD/PCA dimensionality reduction, (iii) document clustering, and (iv) interactive information visualization. This is a high-skill PHOSITA, which increases the likelihood that the mathematical and architectural combinations were routine.

---

3. The prior-art corpus and its § 102 status

Sources: https://patents.google.com/patent/US6775677/en (citation lists); https://patents.google.com/patent/[US5857179A](/patent/US5857179A)/en ; https://patents.google.com/patent/[US5937422A](/patent/US5937422A) ; https://patents.google.com/patent/US5794178 ; https://dl.acm.org/doi/10.1145/[312624](/patent/312624).[312652](/patent/312652) (Ding) ; https://www.semanticscholar.org/paper/a37e61f6be2de0d26c86f315af769387df588f8c (Ding abstract).

All six primary references are in the same field of endeavor (organizing, retrieving, summarizing, and visualizing document collections) or are "reasonably pertinent" to the problem the inventor faced (helping a user understand a set of documents). This supports combinability.

---

4. Claim 1 (and its method/PFP/program-product twins 14, 15, 16)

Claim 1 is a six-element pipeline. The two elements that carry patentable weight are the cosine-based topic-document relevance and the topic sentence set. Each element is addressed by the cited art:

4a. Primary combination: R2 (Vaithyanathan '179) + R1 (Caid '178) + N1 (Ding)

Element mapping:

• "basic term set ... one or more words" → R2 discloses representing each document by an M-dimensional vector over the terms/words of a domain, with a term-by-document matrix. R3 likewise discloses stemming input words to root form, and filtering "stop words" and "lexical collocations." The '677 spec's own term-set step ("extract terms ... remove common terms using a stop word list") is admitted routine.
• "document vector ... direction representing what the document is about" → R2: "each document is represented by a respective M dimensional vector," then "the dimensionality of the initial matrix is reduced to form resultant vectors of the documents." N1: document vectors in the reduced k-dim LSI subspace; and N1 expressly notes that when document vectors are normalized, "cosine similarity is equivalent to dot-product similarity," i.e., direction carries meaning.
• "topic vector process ... topic vectors from the document vectors" → R2's K-means clustering produces cluster centroids; the '677 specification itself says the alternative embodiment of its topic-vector process "is the existing clustering algorithm that takes vectors as input" and "the topic vectors can be created to be the center vectors of the resultant clusters." That is a direct, self-identified equivalence to R2/R1 clustering. R1 also discloses clustering summary vectors and computing a centroid for each cluster.
• "topic term set ... terms describing the topic, each associated with relevancy" → R2 expressly derives, per cluster, the "terms having greatest impact on the documents in that cluster," which "represent key words." R3 derives a "topical description" = "the top N scoring words." Both attach scores (relevance) to the terms.
• "topic-document relevance ... cosine between the topic vector and the document vector" → This is the narrow "how" limitation present in every independent claim. R1 discloses retrieval by comparing vectors "in order of vector proximity" (dot product). N1 explicitly frames LSI around "the cosine similarity measure" and proves cosine ≡ dot product under normalization. A PHOSITA applying LSI (which R2 already does via SVD) would use the cosine measure as a matter of course.
• "topic sentence set ... topic sentences ... associated with relevance" → See the gap analysis in § 8 — this is the weakest-supported element and should be treated as the fulcrum of any validity dispute.

Articulated motivation to combine (KSR):

1. Same problem, complementary solutions. R2 solves "provide keywords/cluster summaries"; R1 solves "visualize relationships among documents." The '677 specification itself criticizes R2-class references ("the labels alone do not provide enough information") and R1-class references ("the information provided by document visualization is too coarse-grained"). The inventor's own framing thus supplies the design incentive to combine a labeled-cluster generator (R2) with a visualizer (R1).
2. Predictable result. Adding relevance-scored term labels to clusters and displaying them is "the predictable use of prior art elements according to their established functions" (KSR). No new, unexpected interaction is required for claim 1 as a whole.
3. N1 supplies the mathematical bridge. Once a PHOSITA has R2's term/document matrix + SVD reduction, using cosine similarity to score topic-vs-document relevance is the baseline convention of the vector-space model (N1 documents this as known).

4b. Optional secondary combination including R3

R2 + R3 independently supports the "topic term set with relevancy" element: R3's "top N scoring words as the topic description" and its "comparing the similarity of the resulting topical descriptions" to search/sort documents. R2 + R3 + N1 would therefore also teach a relevance-ranked topic-term pipeline.

4c. R4 (Light '888)

R4 ("characterizing a document set using evaluation surrogates / topic profiles") is background art that reinforces "characterizing a document set" as a known objective, though it requires predefined topic profiles (a point of distinction the '677 spec exploits). It is best used as corroborating the field and motivation, not as the primary reference.

Bottom line on claim 1: Obvious over R2 + R1 + N1 (with R3 as a supplementary reference for the topic-term element), subject to the sentence-set caveat in § 8.

---

5. Dependent claims 2–4 (the conversion-matrix / interpretation-matrix core)

These are the technically meatiest claims, but they are heavily anticipated-by-convention.

• Claim 2 (term-document vectors + conversion matrix + document-vector conversion + term-sentence vectors + sentence vectors + interpretation matrix):
  • R2 discloses precisely this structure: an "initial document by term matrix," dimensionality reduction producing "a working matrix of loading vectors," and projecting the initial matrix onto the loading vectors ("project initial matrix onto working matrix of loading vectors to generate matrix of reduced dimensionality"). The loading-vector matrix is the "conversion matrix"; the projection is the "document vector conversion."
  • N1 supplies the SVD/LSI math (left singular vectors U_k, right singular vectors V_k, truncation to k) underpinning the conversion and interpretation matrices, and the decoding relation.
  • The interpretation matrix = transpose of the conversion matrix is, in R2's own words, the inverse operation R2 performs to recover term weights: "for each resultant vector, representing each component part of the resultant vector as a sum of products... multiplying a component part of the initial vector and the absolute value of a corresponding component part of a single Eigenvector... summed across all Eigenvectors." That is the transpose/decode step.
  • Sentence vectors (treating a sentence as a small document): the '677 background admits that sentence-level summarization/extraction was known, and applying the vector-space model at the sentence level is a routine, predictable extension of R2/N1. This is the most challengeable sub-element, but under KSR a known technique (vector-space scoring) applied to a known unit (the sentence) is obvious.
• Claim 3 (topic terms = large products of interpretation matrix × topic vector): R2's sum-of-products eigenvector decoding of a cluster centroid into ranked term weights is the same computation. N1's LSI term-space decoding supports it. This claim is highly likely obvious over R2 alone (alone) or R2 + N1.
• Claim 4 (topic sentences selected by sentence-vector magnitude + direction similarity to topic vector): this combines claim 2's sentence vectors with the disclosed cosine/similarity machinery and the admitted known sentence-extraction summarization. Moderate obviousness strength — the specific "large magnitude AND large direction similarity" selection rule is a routine ranking heuristic, but no cited reference expressly teaches operating it on sentence vectors, so this claim has more § 103 daylight than claims 2–3.

---

6. Dependent claims 5–8 (the "document map" visualization)

• Claim 5 (document map; each document an image object positioned and colored by topic-document relevance) → R1 (Caid '178) discloses this almost element-for-element. R1 teaches translating context vectors "into visual and graphical representations," with "Icons represent[ing] specific data objects" and "Icon Position and Attributes (size, color, etc.) represent relationship to coordinate and information words," plus "Coordinate Words specify frame of reference for display." Claim 5's positioned + colored dots are the established function of R1's icons. Strongly obvious over R1 (and R1 taken with R2/N1 for the relevance values being plotted).
• Claim 7 (indicator image object showing the direction of greater relevance) → R1's "coordinate words" / axes establishing a "frame of reference" and FIG. 6/8–9 navigation displays provide this. Obvious.
• Claim 8 (full text displayed when the document image object is selected) → R1 discloses "open icon" behavior (FIG. 9C, "sample display showing an open icon") and interactive navigation/retrieval; R5 (HP '533) discloses the general "method for selecting an item on a graphics screen" via a pointing device. Combination is routine UI design. Obvious.
• Claim 6 (multiple maps; hover over an object changes its color in all maps) → This "linked brushing across multiple coordinated views" feature is the most specific UI limitation in the set, and I could not identify express disclosure of cross-map hover-syncing in the cited references. This claim has the best non-obviousness argument. (Note: coordinated-views/linked-brushing techniques were being published in the information-visualization literature in the late 1990s, but I will not fabricate a specific § 102 antecedent — this is a genuine gap to verify against full-text of R1 and broader art.)

---

7. Dependent claims 9–13 (topic sentence frame + interaction)

• Claim 9 (a topic sentence frame per topic) and claims 10–13 (map + sentence frame displayed together; hover/click on a sentence's image object highlights/opens the source document) are display-layer features.
• R1 supplies the interactive icon framework; R5 (HP '533) supplies item selection on a graphics screen; the '677 spec's sentence frame is a straightforward UI restatement of the same selection/highlight interaction. Likely obvious as an application of known selection/highlight UI to the disclosed data, though the evidentiary support is thinner than for claims 5/7/8.

---

8. The critical weakness — the "topic sentence set" element

Every independent claim (1, 14, 15, 16) requires "a topic sentence set for each topic vector that comprises... topic sentences that describe the topic... each associated with relevance." None of R1–R6 or N1 expressly generates sentences as topic descriptions; R3 generates a word list, not sentences.

However, § 103 (not § 102) governs, and the gap is bridgeable:

• The '677 specification admits that "existing text summarization methods are mainly for single documents... generally based on sentence extraction or sentence generation," and that sentence extraction is a known technique. Under KSR, the known technique of sentence extraction can be combined with the disclosed topic-vector machinery to yield topic sentences — which is exactly what claim 1 asks for when the sentence is scored against the topic vector (the scoring mechanism being R1/N1's cosine).
• Still, the express "topic-sentence-selected-by-vector" link is the element an examiner would most likely need an additional reference for (e.g., a known multi-document/sentence-summarization reference). I flag this as the pivotal element and do not claim the cited set alone conclusively renders it obvious — with only the citations on the patent's face, claim 1's sentence-set limitation is the strongest non-obviousness hook.

---

9. Secondary considerations (objective indicia)

No evidence of secondary considerations appears on the patent's face or the citations list (no unexpected results asserted relative to R1/R2/N1; no licensing/nexus evidence found in the searched material). Issuance of the patent is not, by itself, evidence of non-obviousness. Commercial-success or long-felt-need evidence, if any exists, would need to be produced and tied by nexus; nothing here supplies it. This factor is neutral-to-adverse to patentability on the present record.

---

10. Overall conclusion

The strongest single combination to invalidate the independent claims is US 5,857,179 (Vaithyanathan '179) in view of US 5,794,178 (Caid '178) and Ding (SIGIR '99), with US 5,937,422 (Nelson '422) supplying the relevance-ranked topic-term element — motivated by the shared, admitted problem of helping users comprehend a set of documents and the complementary, predictable functions of the three references (cluster keyword generation + visualization + LSI/cosine scoring).

Honest caveats (do not overstate)

1. My element mappings rely on the abstracts/full text I could retrieve; claims 2 and 4 in particular should be re-checked against the complete R2 and N1 specifications before any formal contention is asserted.
2. The topic-sentence element is not clearly disclosed by the cited references; a stronger § 103 case on claims 1/4/14/15/16 would require at least one additional sentence-summarization reference.
3. Claim 6 (cross-map hover color-syncing) has no express antecedent I could locate in the cited set and is the claim most resistant to the combination above.
4. The patent is expired (2020-03-02), so this analysis is academic/historical unless a retrospective validity question is posed.