Searches and surfaces quotes attributed to entities. Powers the entity-quote SERP feature — the system that pulls verified quotes from a person, organization, or work and presents them inline.
Patent Overview
- Inventor
- Yossi Matias, others
- Assignee
- Google LLC
- Filed
- 2015
- Granted
- 2019-02-05
The Challenge
The Challenge
Users searching for quotes from a public figure want to find verified, attributed quotes. The system needs to identify quotes in source documents, attribute them correctly, verify authenticity, and surface them as a Knowledge Panel feature or quote box.
- Quotes Are Often Misattributed — Famous quotes circulate misattributed. Verification matters.
- Quote Identification Requires NLP — Identifying quoted text vs paraphrased text vs commentary requires NLP analysis.
- Attribution Requires Entity Linking — Per quote, attribution to the right entity requires NER + entity linking.
- Verification Across Sources Builds Confidence — Per quote, multiple-source confirmation builds attribution confidence.
- Surfacing Format Matters — Quote boxes need format (the quote, the attribution, the source) consistent with SERP design.
Innovation
How The System Works
The system identifies quoted text in source documents, attributes each quote to the right entity via NER and linking, verifies attribution across sources, builds a quote database indexed by entity, and surfaces verified quotes on relevant queries.
- Identify Quoted Text — Per document, NLP identifies quoted text vs paraphrased text vs commentary.
- Attribute Quote To Entity — Per quote, NER and entity linking attribute to the right entity.
- Verify Across Sources — Per quote, multi-source corroboration builds attribution confidence.
- Index By Entity — Per (entity, quote), entry in quote database.
- Receive Query — On entity or quote-related query, retrieve quotes.
- Score And Select — Per query, top quotes selected by relevance and verification confidence.
- Surface In SERP — Quote box or Knowledge Panel quote feature rendered.
Verified Quotes Build Trust
The patent's load-bearing idea is that quote surfaces require verification. Identification, attribution, multi-source verification combine to surface only quotes with confidence in authenticity.
Multi-Source Verification Beats Single-Source
Per quote, single-source attribution can be wrong. Multi-source corroboration builds confidence in authenticity. The verification layer is what makes quote surfacing trustworthy.
- Quote Identification — Per document, NLP distinguishes quoted text from other forms.
- Entity Attribution — Per quote, attribution via NER and entity linking.
- Multi-Source Verification — Per quote, multi-source corroboration builds confidence.
Technical Foundation
Technical Foundation
The patent specifies the quote identifier, attribution layer, verifier, quote database, query handler, scorer, and SERP renderer.
- Quote Identifier — Per document, NLP identifies quoted text.
- Attribution Layer — Per quote, NER and entity linking attribute to entity.
- Verifier — Per quote, multi-source verification builds confidence.
- Quote Database — Per (entity, quote), persistent entry indexed by entity.
- Query Handler — Per query, quote retrieval and scoring.
- SERP Renderer — Surfaces selected quotes in quote box or Knowledge Panel.
The Process
The Process
Quote identification runs at indexing; query-time retrieval and selection at query time.
- Index Documents — Documents crawled and quoted text identified.
- Attribute Quotes — Per quote, attribution via NER and linking.
- Verify Across Sources — Multi-source corroboration.
- Index By Entity — Per (entity, quote), database entry.
- Receive Query — Entity or quote query arrives.
- Retrieve And Score — Quotes retrieved and scored.
- Render — Top quotes surfaced.
Quality Control
Quality Control
Quote misattribution damages trust. The patent specifies safeguards.
- Multi-Source Verification Threshold — Per quote, multi-source corroboration required above threshold.
- Attribution Confidence Validation — Per attribution, confidence validated.
- Source Quality Filtering — Per source, source quality required. Low-quality sources don't contribute.
- Adversarial Defense — Misattributed-quote propagation patterns flagged and filtered.
- Continuous Recalibration — Identifier, attribution, verifier models recalibrate against fresh data.
Real-World Application
Quote search underpins the entity-quote SERP feature and Knowledge Panel quote sections. The identification-attribution-verification pattern is foundational for entity-content trust.
- NLP-identified Quote Detection — Per document, NLP distinguishes quotes from other text forms.
- Entity-linked Attribution — Per quote, NER plus entity linking attribute to right entity.
- Multi-source Verification — Multi-source corroboration builds attribution confidence.
Why Properly Attributed Citations Win
Per quote, source-page attribution must be clear and verifiable. Pages with clear citation patterns (named speaker, citation source, original date) contribute high-confidence quote-database entries.
Why Original Source Documentation Compounds
When your content is the original source of a quote, properly documenting that establishes you as the canonical reference. Multi-source verification favors well-documented originals.
<\/section>What This Means for SEO
What This Means for SEO
This patent identifies quotes in documents, attributes them to entities via NER and entity linking, and verifies attribution across multiple sources before surfacing them. SEO implication: clearly attributed citations and well-documented original sources earn high-confidence quote-database entries.
- Attribute Citations Clearly And Verifiably — Quote attribution depends on clear, verifiable source pages. Pages naming the speaker, citation source, and original date contribute high-confidence quote-database entries, where vague attribution does not.
- Document Original Sources To Become Canonical — When your content is the original source of a quote, documenting that establishes you as the canonical reference. Multi-source verification favors well-documented originals over later repeaters.
- Multi-Source Verification Builds Confidence — A quote needs corroboration across multiple sources above a threshold before surfacing. Single-source claims are treated cautiously, so quotes confirmed across credible sources earn surfacing.
- Clear Quote Formatting Aids Identification — NLP distinguishes quoted text from paraphrase and commentary. Presenting quotes with clear quotation formatting and explicit attribution helps the identifier recognize them correctly.
- Entity Linking Requires Clean References — Attribution runs through NER and entity linking. Referring to the quoted entity by a clear, consistent, identifiable name helps the system attribute the quote to the right entity rather than a namesake.
- Misattribution Patterns Are Filtered — Misattributed-quote propagation patterns are flagged and filtered. Repeating popular but unverified attributions does not build standing; accurate, sourced attribution does.
- Source Quality Gates Contribution — Low-quality sources do not contribute to verification. Earning quote-database standing requires being a credible source, so authority and clear documentation work together.