Guide : Semantic Search

TL;DR — Semantic search signifie moteur de recherches match the meaning of ce que you’re looking pour, pas simplement the exact words vous typed. Google figures out “what does this person actually want?” and “what is this page really about?” — alors connects the two. It’s pourquoi vous don’t have to repeat keywords over and over anymore, and pourquoi writing naturally à propos de a topic fonctionne meilleur que stuffing it with phrases.

Ce que semantic search is

Semantic search seeks matches fondé on meaning and context en outre to exact terms. Evidence for this claim BERT learns bidirectional contextual language representations that can be fine-tuned for search-relevant language tasks. Scope: BERT research; semantic retrieval systems may use many other models and signals. Confidence: high · Verified: 2026-07-14 Devlin et al.: BERT Modern systems may combine language models, lexical retrieval, entities, and autre signals plutôt que rely on un technique. Evidence for this claim Google reported using BERT to better understand language and context in some Search queries. Scope: Google Search's documented rollout; it does not mean lexical matching was replaced or disclose the full ranking system. Confidence: high · Verified: 2026-07-14 Google: Understanding searches better than ever before

Pour a long temps, moteur de recherches mostly counted words. Si vous searched “cheap flights,” they looked for pages with the words “cheap” and “flights” on les. Utile, but dumb — it had aucun idea que “affordable airfare” signifie the même chose, or que “jaguar” pourrait be an animal or a car.

Semantic search fixes que. Au lieu de matching letters and words, it tries to comprendre meaning. Two ideas drive it:

• Understanding ce que vous mean. Google reads votre whole requête as a sentence, pas a bag of keywords. It fonctionne out votre intent — are vous trying to buy something, apprendre something, or trouver a spécifique site?
• Understanding ce que une page is à propos de. It figures out the topics une page covers and the real-world choses (personnes, places, products) it mentions — pas simplement qui words apparaître.

A bon façon to picture it: searching “how to fix a leaky faucet” and “repairing a dripping tap” devrait retourner the même results, même though ceux phrases share almost aucun words. To a keyword engine ils regarder unrelated. To a semantic engine ils mean the même chose.

Pourquoi it modifié SEO

Quand Google understood meaning, a lot of old tricks stopped working:

• Vous don’t besoin to stuff keywords. Google déjà knows synonyms and connexe terms. Repeating “best running shoes” twenty times doesn’t aider — it peut hurt.
• Vous don’t besoin a separate page pour every wording. “Python tutorial,” “Python guide,” and “apprendre Python” tout mean roughly the même chose, so Google picks un page to rank pour tout of les. Splitting les à travers pages simplement competes with yourself.
• Covering a topic bien beats hitting a keyword count. Une page que genuinely réponses a question — with the connexe details a reader voudrait vouloir — outperforms un que simplement mentions the keyword a lot.

The chose la plupart personnes obtenir incorrect

Semantic search ne fait pas mean keywords are dead. Keyword research encore indique vous ce que personnes search pour and how souvent. Ce que died is keyword stuffing and the idea que vous doit repeat exact phrases. And to be clair: “LSI keywords” ne sont pas a réel chose. Google’s John Mueller said so directement. Sprinkling “semantically connexe keywords” from a outil isn’t how ce fonctionne — covering the topic naturally is.

Vouloir the complet history (Knowledge Graph, Hummingbird, RankBrain, BERT, MUM) and the technical detail of Comment cela fonctionne? Switch to the Avancé tab.

TL;DR — Semantic search retrieves by meaning, pas exact keyword match. Google construit it in layers — Knowledge Graph (2012), Hummingbird (2013), RankBrain (2015), neural matching (2018), BERT (2019), MUM (2021) — chaque solving a différent piece (entities, whole-query meaning, novel requêtes, concept-level synonyms, contextual word meaning, multimodal reasoning). Ils augment keyword retrieval (BM25 encore runs premier at scale), ils don’t replace it. Semantic search is the goal; vector search is un implementation. “LSI keywords” are a myth. The optimization réponse hasn’t modifié in a decade: natural language, topical depth, clair entities, intent alignment.

Keyword search vs. semantic search

Keyword and semantic retrieval are complementary, pas mutually exclusive eras. Evidence for this claim BERT learns bidirectional contextual language representations that can be fine-tuned for search-relevant language tasks. Scope: BERT research; semantic retrieval systems may use many other models and signals. Confidence: high · Verified: 2026-07-14 Devlin et al.: BERT Google’s public explanations confirmer language-understanding systems sans revealing complet ranking weights. Evidence for this claim Google reported using BERT to better understand language and context in some Search queries. Scope: Google Search's documented rollout; it does not mean lexical matching was replaced or disclose the full ranking system. Confidence: high · Verified: 2026-07-14 Google: Understanding searches better than ever before

Classic retrieval ranks documents by term statistics. TF-IDF (term frequency × inverse document frequency) weights a word by how souvent it apparaît in a document contre how rare it is à travers the corpus. BM25 (“Best Match 25”) refines que with length normalization and a saturation curve, and it’s encore the dominant first-stage baseline in Elasticsearch, Solr, and Lucene — and à l’intérieur Google and Bing. Powerful, fast, and entirely à propos de words. It has aucun idea que “leaky faucet” and “dripping tap” mean the même chose.

Semantic search adds a couche of understanding on top:

The important framing: semantic search augments keyword retrieval plutôt que replacing it. At Google scale, an inverted-index BM25-style réussir encore fait the high-recall premier cut pour speed, alors neural systems re-rank and ajouter documents que share concepts but pas words. Pandu Nayak’s 2023 DOJ testimony décrit exactly ce — an embedding system (internally RankEmbed) que “identifies a few plus documents to ajouter to ceux identified by the traditional retrieval.” Keyword retrieval didn’t die; it got a meaning-aware partner.

How Google construit semantic search — the milestones

The evolution was incremental. Aucun unique mettre à jour flipped Google from “keywords” to “meaning” — it was a decade of layers, chaque solving a distinct problem.

2012 — Knowledge Graph (“things, not strings”)

Google’s database of real-world entities and leur relationships. Amit Singhal’s launch framing — “things, not strings” — is encore the unique meilleur one-line explanation of the semantic shift. It launched with 500+ million entities and 3,5+ billion facts; the framing matters plus que the numbers. Ce is ce que lets Google disambiguate “Taj Mahal” (monument, musician, or casino) and réponse entity questions directement. The Knowledge Graph is the entity backbone the rest of the stack leans on — I go deeper on it in entity SEO.

2013 — Hummingbird (whole-query meaning)

A complet rewrite of Google’s core requête engine — pas a tweak comme Panda or Penguin, but a nouveau engine. Singhal appelé it the la plupart dramatic modifier since 2001. The goal was conversational and long-tail requêtes: paying attention to the whole requête — the whole sentence, the meaning — plutôt que particulier words. Danny Sullivan’s contemporaneous summary captures it bien: Hummingbird “is paying plus attention to chaque word in a requête, ensuring que the whole requête… is taken into account, plutôt que particulier words.” Google dit it affected 90% of searches. (It’s now listed as retired in the Ranking Systems Guide — superseded by the systems que evolved out of it.)

2015 — RankBrain (never-before-seen requêtes)

Machine learning applied to requête interpretation. RankBrain’s job is the roughly 15% of daily requêtes Google had jamais seen avant — it converts an unfamiliar requête into a mathematical vector and trouve conceptually similaire requêtes it fait comprendre. Greg Corrado, the Google scientist who confirmed it (via Bloomberg, pas a Google blog post), décrit it as embedding “vast amounts of written language into mathematical entities — appelé vectors — que the computer peut comprendre.” By 2016 it processed every requête. Remarque the boundary: RankBrain maps unknown requêtes to connu concepts; it’s pas the même as BERT.

2018 — Neural matching (concept-level synonyms)

Où RankBrain connexe requêtes to concepts, neural matching extended concept-matching to the document side — connecting a query’s concepts to une page’s concepts même quand ils share aucun vocabulary. Danny Sullivan’s plain-language description is the meilleur un out là: “Dernier few months, Google has been en utilisant neural matching, [an] AI méthode to meilleur connecter words to concepts. Super synonyms, in a façon, and impacting 30% of requêtes.” The canonical example: “pourquoi fait my television regarder strange” peut surface results à propos de the “soap opera effect” — a concept neither phrase noms. (Internally, the 2023 DOJ testimony revealed ce lineage as RankEmbed / RankEmbedBERT.)

2019 — BERT (contextual word meaning)

The big NLP breakthrough. BERT (Bidirectional Encoder Representations from Transformers) reads a word in context — looking at the words avant and après it, plutôt que left-to-right un at a temps. That’s ce que lets it catch how “to” changements the meaning of “2019 brazil traveler to usa need a visa” (a Brazilian traveling to the US, qui older systems got backwards). Pandu Nayak appelé it the “biggest leap forward in the past five années, and un of the biggest leaps forward in the history of Search,” affecting 1 in 10 US English requêtes at launch and now nearly tout of les. Ce is aussi the mettre à jour SEOs la plupart souvent misread: Danny Sullivan’s réponse was blunt — “There’s nothing to optimize for with BERT.”

2021 — MUM (multimodal, multilingual)

Multitask Unified Model — construit on the T5 framework, trained à travers 75 languages, and able to comprendre information à travers text and images. Google dit it’s 1 000× plus powerful que BERT and, unlike BERT, peut les deux comprendre and generate language. Important caveat: MUM is utilisé pour spécifique high-complexity cas (complex multi-step questions, certain featured snippets, shopping) — it was jamais Google’s general ranking engine, and it didn’t “replace” BERT.

2023+ — LLMs and AI search

Semantic retrieval now feeds generative réponses. AI Overviews and AI Mode utiliser the même meaning-based retrieval to trouver relevant passages, alors an LLM synthesizes an réponse — voir RAG. The semantic couche determines ce que obtient retrieved and cited; the model simplement writes it up.

Semantic search vs. vector search

Ces obtenir blurred constantly, so be precise:

• Semantic search is the goal — retrieve by meaning and intent.
• Vector search is un implementation — represent requêtes and documents as dense vectors and trouver nearest neighbors by cosine distance.

Vector search is the la plupart courant modern implementation, but it isn’t the seulement chemin to semantic search: requête expansion, synonym rules, Knowledge Graph lookups, and entity recognition tout obtenir vous là sans computing a unique embedding. Think of semantic search as the destination and vector search as un (fast, scalable) vehicle. The mechanics of the vehicle — embeddings, cosine similarity, nearest-neighbor search — are covered in the embeddings and vector search articles; I won’t re-derive les ici.

How it en réalité fonctionne sous the hood

Four moving parts, roughly in order:

1. Requête understanding — intent classification (informational, navigational, transactional, commercial), entity extraction, and synonym/concept expansion. Ce is RankBrain and BERT territory.
2. Entity recognition + the Knowledge Graph — identifying the real-world choses a requête and a document are à propos de, and disambiguating les. À propos de 40% of English words have multiple meanings; context resolves qui un vous mean.
3. Embeddings and semantic similarity — encoding meaning as vectors so “fix a leaky faucet” and “repairing a dripping tap” land fermer ensemble (~0,89 similarity despite barely sharing a word). Deep dive in embeddings.
4. Passage ranking and neural re-ranking — since 2020, individual passages of a page peut rank pour a requête même quand the whole page isn’t perfectly targeted, and a neural re-ranker reorders the keyword-retrieved définir by semantic fit. Ce is pourquoi chaque section nécessite to stand on its propre.

The LSI keyword myth

Ce un matters parce que it drives a lot of bad advice. “LSI keywords” (Latent Semantic Indexation) is a 1988 information-retrieval technique que Google has jamais confirmed en utilisant as a ranking input. John Mueller put it flatly: “There’s aucun tel chose as LSI keywords — anyone who’s telling vous sinon is mistaken, sorry.” Ce que Google en réalité uses is far plus sophisticated — neural matching, BERT, word embeddings, entity recognition. So quand a outil hands vous a liste of “LSI keywords,” what’s utile à propos de it isn’t the LSI partie; it’s que ceux terms reflect the vocabulary of the topic. Cover the topic naturally and vous obtenir que pour free.

Ce que semantic search signifie pour le SEO

Every major semantic mettre à jour has pushed in the même direction, qui rend the playbook unusually stable:

• Topic coverage beats keyword density. Google judges si une page covers a topic thoroughly, pas si it hits a keyword n times. Deep coverage ranks pour dozens of connexe requêtes vous jamais targeted individually.
• Entity optimization. Nom and décrire votre entities clearly; utiliser structured données (sameAs, @id) to disambiguate les contre Wikipedia/Wikidata. Ce is the entity SEO discipline.
• Intent alignment is non-negotiable. Google classifies requête intent and filters by it. Une page que réponses a différent intent que the requête won’t rank aucun matter how bien the keywords match.
• Écrire naturally. With BERT reading context, prepositions and sentence structure carry meaning. Gary Illyes’ advice on RankBrain encore holds: “Si vous essayer to écrire comme a machine alors RankBrain va simplement obtenir confused and probably simplement pushes vous back.”
• Synonyms are handled pour vous. Vous don’t besoin every variation; covering the topic’s natural vocabulary signals depth sans stuffing.
• Passage-level quality. Chaque H2/section devrait stand alone as a clean, self-contained réponse — that’s ce que passage ranking and AI Overviews retrieve.

Quand semantic SEO matters moins

Honesty vérifier: pour a single-service local business or a thin site, heavy entity and topical-authority fonctionner may pas pay off. The ROI montre up quand you’re competing on informational depth à travers a topic. As Sally Mills put it, “Si vous do SEO correctement, you’re automatically doing semantic SEO. It’s simplement que la plupart personnes aren’t doing it correctement.” And Google won’t go purely semantic anytime soon — complet semantic retrieval is expensive, exact-match is encore courant utilisateur behavior, and purely semantic results remain unreliable. Keyword retrieval and semantic understanding coexist.

How AI search builds on ce

AI Overviews, AI Mode, and AI assistants are semantic search plus generation. Ils retrieve passages by meaning (usually dense retrieval — voir RAG and vector search), alors an LLM writes the réponse. Bing’s framing of the shift is sharp: grounding indexation, in leur words, “is being construit to aider AI systems decide ce que to dire.” The practical consequence: the même choses que faire a passage rank bien — clair entities, self-contained sections, topical depth — are ce que faire it probable to be cited in an AI réponse. There’s aucun separate trick.

AI summary

A condensed prendre on the Avancé version:

• Semantic search = retrieval by meaning, pas exact keyword match. Que distinction is the whole point. “Fix a leaky faucet” and “repairing a dripping tap” devrait retourner the même results despite sharing aucun words.
• It augments keyword search, it doesn’t replace it. BM25-style retrieval encore runs premier at scale (speed + recall); neural systems re-rank and ajouter concept-matched documents. DOJ testimony confirmed Google’s embedding couche (RankEmbed) adds documents to traditional retrieval.
• Google construit it in layers: Knowledge Graph (2012, “things not strings”) → Hummingbird (2013, whole-query meaning, 90% of searches) → RankBrain (2015, novel requêtes) → neural matching (2018, concept-level “super synonyms,” 30% of requêtes) → BERT (2019, contextual word meaning, 1-in-10 requêtes) → MUM (2021, multimodal, 75 languages, jamais the general ranking engine).
• Semantic search ≠ vector search. Semantic search is the goal; vector search is un implementation (others: requête expansion, synonym rules, Knowledge Graph lookups).
• “LSI keywords” are a myth — John Mueller said so directement. Google uses neural matching, BERT, and embeddings à la place.
• SEO playbook (unchanged à travers a decade): topic coverage over keyword density, clair entities + données structurées, intent alignment, natural-language writing, passage-level quality. Keyword research encore matters; keyword stuffing doesn’t.
• AI search = semantic retrieval + generation. Ce que ranks bien (clair entities, self-contained passages, depth) is ce que obtient cited.

Documentation officielle

Primary-source documentation and announcements from Google and Microsoft.

Google

• A guide to Recherche Google ranking systems — official current descriptions of BERT, neural matching, RankBrain, MUM, and Passage Ranking (and the liste of retired systems, notamment Hummingbird).
• Introducing the Knowledge Graph: choses, pas strings — the 2012 entity launch and the “things, not strings” framing.
• Understanding searches meilleur que ever avant (BERT) — Pandu Nayak’s 2019 BERT announcement, with the visa/esthetician exemples.
• MUM: A nouveau AI milestone pour understanding information — the 2021 multimodal, multilingual model.
• How AI powers great résultats de recherche — Google’s overview of how RankBrain, neural matching, BERT, and MUM fit ensemble.

Microsoft / Bing

• The Science Behind Semantic Search (Azure AI) — “a Transformer-based semantic ranking engine that understands the meaning behind the text.”
• Semantic ranking in Azure AI Search — the BM25/RRF retrieval → transformer re-ranking pattern, spelled out.
• Towards Plus Intelligent Search: Deep Learning pour Requête Semantics — rare technical depth on Bing’s pre-BERT neural ranking (GloVe vectors, CNN + cosine similarity).
• Introducing Deep Search — un of the earliest public descriptions of requête fan-out at production scale.

Quotes from the source

On-the-record statements from Google and Microsoft reps. Où a deep lien is disponible it jumps to the quoted passage on the source page.

Google — the semantic shift

• “Things, not strings.” — Amit Singhal, introducing the Knowledge Graph, 2012. Jump to quote

Google — BERT (Pandu Nayak, 2019)

• “BERT models can therefore consider the full context of a word by looking at the words that come before and after it — particularly useful for understanding the intent behind search queries.”
• “[The] biggest leap forward in the past five years, and one of the biggest leaps forward in the history of Search.”
• “Language understanding remains an ongoing challenge, and it keeps us motivated to continue to improve Search.” Jump to quote

Danny Sullivan, Recherche Google Liaison

• On neural matching (2018): “Last few months, Google has been using neural matching — AI method to better connect words to concepts. Super synonyms, in a way, and impacting 30% of queries.”
• “RankBrain helps Google better relate pages to concepts. Neural matching helps Google better relate words to searches. It’s like a super synonym system.”
• On optimizing pour BERT (2019): “There’s nothing to optimize for with BERT, nor anything for a site owner to be ‘concerned’ about. The fundamentals of us seeking to reward great content remain unchanged.”

Gary Illyes, Google

• On writing pour RankBrain (Big Digital Adelaide, 2016): “Optimizing for RankBrain is actually super easy… write in natural language. Try to write content that sounds human. If you try to write like a machine then RankBrain will just get confused and probably just pushes you back.”

John Mueller, Google

• On the LSI myth (Twitter, July 30, 2019): “There’s no such thing as LSI keywords — anyone who’s telling you otherwise is mistaken, sorry.”

Greg Corrado, Google (on RankBrain, via Bloomberg, 2015)

• “RankBrain uses artificial intelligence to embed vast amounts of written language into mathematical entities — called vectors — that the computer can understand. If RankBrain sees a word or phrase it isn’t familiar with, the machine can make a guess as to what words or phrases might have a similar meaning and filter the result accordingly.”

Pandu Nayak, Google (DOJ antitrust testimony, 2023)

• “RankEmbed identifies a few more documents to add to those identified by the traditional retrieval.” (Confirms que Google’s semantic embedding couche augments classical keyword retrieval plutôt que replacing it.)

Microsoft — Azure AI / Bing

• “Semantic search is moving beyond keyword-based ranking to a Transformer-based semantic ranking engine that understands the meaning behind the text.” — Microsoft Research, 2021.

Google’s semantic milestones — cheat sheet

The layered stack, in order. Chaque solved a différent problem; none replaced the un avant it.

Rapide distinctions to garder straight

• RankBrain vs. BERT — RankBrain maps unknown requêtes to connu concepts; BERT understands the contextual meaning of words dans a requête sentence.
• Semantic search vs. vector search — semantic search is the goal; vector search is un implementation (others: requête expansion, synonym rules, Knowledge Graph).
• Semantic SEO — topic coverage, entity relationships, intent matching. Pas synonym stuffing, and definitely pas “LSI keywords” (a myth — voir the article).
• Keyword search isn’t dead — BM25-style retrieval encore runs premier; semantic systems re-rank and ajouter concept-matched documents on top.

The mental models

1. Goal vs. implementation. Semantic search is the destination (retrieve by meaning). Vector search is un vehicle (embeddings + nearest-neighbor). Vous pouvez aussi walk là — requête expansion, synonym rules, entity lookups. Don’t conflate the destination with un route.

2. The layered stack, pas a switch. Knowledge Graph (entities) + RankBrain (unknown requêtes) + neural matching (concept synonyms) + BERT (word context) + MUM (multimodal). Ils stack and cooperate; aucun mettre à jour threw out the previous un. Quand vous raison à propos de “how Google understands ce requête,” demander qui couche is doing the fonctionner.

3. Retrieval is two-stage. Stage 1: fast, keyword-based, high-recall (BM25 over an inverted index). Stage 2: neural re-ranking + concept-matched additions. Semantic search is mostly stage 2 — it augments stage 1, it doesn’t delete it.

4. The optimization invariant. Every semantic mettre à jour points the même façon: natural language, topical depth, clair entities, intent alignment. Si a tactic seulement rend sense to a keyword-counting engine (stuffing, exact-match obsession, “LSI keywords”), it’s optimizing pour a machine que ne … plus exists.

5. The “what gets cited” rule pour AI search. AI réponses = semantic retrieval + generation. Ce que ranks a passage (clair entities, self-contained sections, depth) is ce que obtient it retrieved and cited. There’s aucun separate AI-search trick — it’s the même passage-level quality.

Semantic-search mistakes

Building pages autour listes of “LSI keywords”

LSI keyword advice mislabels how modern retrieval fonctionne. Cover the subject naturally, nom réel entities and relationships, and réponse the user’s intent au lieu de ajout a mechanically generated synonym liste.

Treating semantic search and vector search as synonyms

Semantic retrieval is the goal of matching meaning; vectors are un méthode que peut prise en charge it. Garder the distinction quand evaluating a system or choosing an audit.

Repeating the head term au lieu de resolving ambiguity

Plus repetition ne fait pas clarify qui entity or intent une page sert. Utiliser explicit noms, attributes, context, and connected subtopics so meaning survives au-delà a unique keyword string.

Testez vos connaissances: Semantic search

Ressources utiles

My connexe reviews

• Semantic SEO (Ahrefs) — the avancé practitioner guide (I’m a reviewer on ce un), covering the EAV model and pourquoi Google won’t go purely semantic.

On ce site

• Entity SEO — the practical discipline of naming, describing, and disambiguating entities (the Knowledge Graph side of semantic search).
• Embeddings and vector search — the technical mechanism que rend meaning-based similarity fonctionner.
• RAG — how AI search retrieves passages and grounds generated réponses.

From others

• Semantic Search — Mateusz Makosiewicz’s Ahrefs guide to meaning-based retrieval, entities, and content.
• Google’s Knowledge Graph — Despina Gavoyannis and Michal Pecánek’s Ahrefs guide to entity données and SEO.
• RankBrain glossary and Hummingbird glossary (Ahrefs) — Référence rapide entries from the Ahrefs editorial team.
• Google’s Search ranking systems guide — the authoritative liste of what’s live and what’s retired.
• SEO by the Sea — Hummingbird retrospective — detailed history with Singhal context.
• Google Hummingbird (Moteur de recherche Land) — Danny Sullivan’s original 2013 announcement report with Singhal quotes; principal source pour “90% of searches” and full-phrase meaning.
• Meet RankBrain, Google’s Artificial Intelligence Search Algorithm (Moteur de recherche Land) — the original 2015 report breaking RankBrain, with Greg Corrado’s verbatim quotes on vectors and unknown requêtes.
• Google Neural Matching (Moteur de recherche Journal) — covers the 2018 neural matching launch, Danny Sullivan’s “super synonyms” tweet, and the distinction from RankBrain.
• FAQ: Tout À propos de the BERT Algorithm dans la recherche Google (Moteur de recherche Land) — comprehensive Q&A notamment Danny Sullivan’s “nothing to optimize for” guidance.
• Google Turning Its Lucrative Web Search Over to AI Machines (Bloomberg) — the original RankBrain disclosure; Greg Corrado confirms it as “the third-most important signal.”
• The Science Behind Semantic Search: How AI from Bing Is Powering Azure Cognitive Search (Microsoft Research) — Bing’s BM25 → transformer re-ranking pipeline explained; inclut the 2% / 4,5% CTR lift données from Microsoft Docs testing.

Videos

• Recherche Google Central (YouTube) — the How Recherche Google Fonctionne series, notamment the explainers on requête understanding and the systems behind it. Channel

Stats worth citing

• ~40% of English words have multiple meanings — qui is pourquoi context (emplacement, history, surrounding words) is ce que semantic search uses to disambiguate “apple” or “jaguar.” Source
• ~0,89 vector similarity entre “how to fix a leaky faucet” and “repairing a dripping tap” — despite sharing almost aucun words. The concrete face of meaning-based matching. Source
• Neural matching affecte ~30% of requêtes (Danny Sullivan, 2018) — the scope of concept-level “super synonym” matching.
• BERT affected 1 in 10 US English requêtes at launch (Pandu Nayak, 2019) and now plays a role in nearly every English requête. Source
• The Knowledge Graph holds ~1,6 trillion facts à propos de ~54 billion entities today (up from 500M entities at 2012 launch) — the scale of Google’s entity backbone. Source

Journal des modifications