GlossaryCopy Anchor Link

A plain-language dictionary for AACSearch customers, grouped by area. Terms marked with an asterisk (*) are specific to AACSearch or to how this platform uses the concept.

1. Search foundationsCopy Anchor Link

• Index — a searchable set of documents with a defined schema; the unit you query.
• Collection — the underlying store that backs an index (collection + schema + settings).
• Document — a single record in an index: a product, a page, an article.
• Field — a named attribute of a document, e.g. title, price, brand.
• Schema — the declared list of fields and their types for an index.
• Tokenization — splitting text into the tokens (words) that get indexed.
• Token — one indexed unit of text, usually a word.
• Inverted index — the structure mapping each token to the documents containing it.
• Stemming — reducing words to a shared root (running → run).
• Lemmatization — reducing a word to its dictionary form using morphology.
• Stop words — very common words ignored at index/query time (the, and, a).
• Synonym — a word mapped to equivalent terms (sneaker = trainer).
• N-gram — a sequence of n characters/words, used for partial and typo matching.
• Prefix search — matching terms by their beginning (pho → phone).
• Suffix search — matching terms by their ending.
• Inflection — grammatical word forms (singular/plural, case).
• Phonetic matching — matching by sound (e.g. names and brands).
• Language detection — detecting the query language to apply the right rules.
• Character normalization — folding diacritics and case (é → e).
• Query — the text a user searches for.
• Query terms — the individual tokens after a query is parsed.
• Search request — one call to the search API or widget.
• Search result — one document returned for a query.
• Hit — a result that matched; often used loosely for "result".
• Result set — the ordered list of returned documents.
• Pagination — splitting a result set into pages.
• Page / per_page — the requested page number and page size.
• Sort — ordering results by a field instead of relevance.
• Grouping — collapsing near-duplicate results (e.g. same product, many variants).
• Distinct — removing duplicate values of a field from the result.
• Highlighting — marking the matched words inside a returned snippet.
• Snippet — a short, highlighted excerpt shown under a result.
• Wildcard — * or ? matching unknown characters in a query.
• Regex query — pattern-based matching using regular expressions.
• Escaping — treating special characters as literal text.

2. Relevance and rankingCopy Anchor Link

• Relevance — how well a result matches the query intent.
• Ranking — ordering results by computed relevance.
• Relevance score — the numeric score assigned to each result.
• Keyword match — matching by shared tokens between query and document.
• Text match level — how strongly tokens align: exact, strong, weak, none.
• Exact match — the query tokens match a field exactly.
• Partial match — only some tokens match.
• Field weighting — giving some fields more influence on the score.
• Boosting — increasing a field or value's contribution to the score.
• Burying — decreasing a document's score so it ranks lower.
• Pinning — forcing a specific document to a fixed position.
• Ranking rule — a declarative adjustment: pin, boost, bury, synonym.
• Search profile* — a saved set of ranking settings applied to queries.
• Default ranking — the ranking used when no profile is active.
• Tie-breaking — a deterministic rule when two results score equally.
• Score decay — reducing the score of older or less important matches.
• Popularity signal — using clicks/sales to influence ranking.
• Freshness — using recency to influence ranking.
• Proximity — rewarding query terms that appear close together.
• Term frequency (TF) — how often a term occurs in a document.
• Inverse document frequency (IDF) — how rare a term is across the index.
• TF-IDF — a classic weighting combining TF and IDF.
• BM25 — a modern probabilistic ranking formula.
• Okapi BM25 — the most common BM25 variant.
• k1 / b — BM25 tuning parameters (term saturation and length normalization).
• Length normalization — reducing the score of very long fields.
• Boosting function — a formula or multiplier applied to the score.
• Query-time boosting — boosts applied only for a specific query.
• Index-time boosting — boosts baked in at indexing time.

3. Search UX and matchingCopy Anchor Link

• Autocomplete — suggestions as the user types each keystroke.
• Typeahead — another name for autocomplete.
• Query suggestion — "did you mean" corrections.
• Spell correction — automatically fixing likely typos.
• Typo tolerance — matching despite typos (ipone → iphone).
• Fuzzy matching — tolerance for small spelling differences.
• Edit distance — how many character edits separate two words.
• Facet — an aggregated filter with counts (brand, price, category).
• Faceting — computing facet counts for a result set.
• Filter — a hard constraint excluding non-matching documents.
• Facet value — one value inside a facet (e.g. brand = Acme).
• Numeric range — a filter over a numeric interval (price).
• Multi-select facet — choosing several facet values at once.
• Instant search — results on every keystroke, no submit step.
• Search-as-you-type — another name for instant search.
• Geo search — ranking or filtering by location.
• Bounding box — a rectangular geographic filter.
• Radius search — a geographic filter by distance from a point.
• Zero results — a query that returned nothing.
• No-results page — the UI state shown on zero results.
• Related queries — similar queries suggested on a results page.
• Query history — the user's past queries.
• Popular queries — the most frequent queries in an index.
• Trending queries — queries rising in frequency recently.
• Search bar — the input element where users type queries.
• Search overlay — a full-screen search UI opened over the page.
• Voice search — search driven by spoken input.
• Image search — search driven by an image instead of text.
• Filters UI — the visible controls for facets and filters.

4. Metrics and analyticsCopy Anchor Link

• Precision — fraction of returned results that are relevant.
• Recall — fraction of all relevant documents that were returned.
• F1 score — the harmonic mean of precision and recall.
• Mean reciprocal rank (MRR) — a ranking-quality metric.
• Normalized DCG (nDCG) — a graded ranking-quality metric.
• Zero-result rate — share of queries returning no results.
• CTR (click-through rate) — share of searches with at least one click.
• Add-to-cart rate — share of searches leading to an add-to-cart.
• Conversion rate — share of searches leading to a purchase.
• Revenue per search — revenue divided by the number of searches.
• Session — one user's continuous interaction.
• Query log — the stored record of queries and their outcomes.
• Search analytics — metrics derived from the query log.
• Funnel — the sequence search → click → add-to-cart → purchase.
• Attribution — assigning revenue/actions to the source (e.g. a query).
• Baseline — the starting score before an experiment.
• Uplift — the measured improvement over baseline.
• A/B experiment* — splitting live traffic between two search profiles.
• Control arm — the unchanged group in an experiment.
• Treatment arm — the changed group in an experiment.
• Sticky assignment — keeping a visitor in the same arm across sessions.
• Statistical significance — whether a difference is unlikely to be noise.
• Golden set — a curated set of queries with expected good results.
• Evaluation set — queries used to score a ranking change.
• Click model — a model of relevance learned from clicks.
• Engagement — how users interact with results.
• Bounce — a visitor leaving after one page/action.
• Retention — whether users return over time.
• Cohort — a group of users sharing a start event.
• AOV (average order value) — average revenue per order.
• LTV (lifetime value) — total value a customer brings over time.
• Churn — customers who stop using the product.
• NPS (net promoter score) — a satisfaction metric.
• CSAT — customer satisfaction score.
• Latency — the time a request takes.
• p50 / p95 / p99 latency — latency percentiles.
• Throughput — requests served per second.
• Error rate — the share of failing requests.

5. Semantics and vectorsCopy Anchor Link

• Semantic search — matching by meaning, not just tokens.
• Embedding — a fixed-length vector representing meaning.
• Vector — an ordered list of numbers (here, a meaning representation).
• Vector search — nearest-neighbour search over embeddings.
• Hybrid search — combining keyword and vector signals.
• Alpha — the blend weight between keyword and vector in hybrid ranking.
• Dense vector — a vector where most values are non-zero (typical embeddings).
• Sparse vector — a mostly-zero vector (e.g. lexical/BM25 weights).
• Reranking — a second pass reordering top candidates with a stronger model.
• Distance metric — how closeness between vectors is measured.
• Cosine similarity — a distance metric measuring angle between vectors.
• Euclidean distance — straight-line distance between vectors.
• Dot product — a distance metric summing element products.
• ANN (approximate nearest neighbour) — fast approximate vector search.
• HNSW — a common graph-based ANN index structure.
• IVF — an ANN structure partitioning vectors into clusters.
• Recall@k — whether the correct result is in the top k.
• kNN — k nearest neighbours.
• Auto-embedding* — the engine computing vectors from fields you point at.
• Explicit vectors — vectors you compute and send yourself.
• Vector field — a float[] field holding an embedding.
• Model — the neural model that produces embeddings.
• Fine-tuning — further training a model on specific data.

6. AI and RAGCopy Anchor Link

• AI answers — answers generated over retrieved content.
• Assistant — the conversational AI search interface.
• LLM (large language model) — a model that generates text.
• RAG (retrieval-augmented generation) — answering from retrieved passages with citations.
• GraphRAG — RAG grounded in a knowledge graph.
• Retrieval — finding relevant passages for a question.
• Generation — writing the answer from retrieved passages.
• Chunking — splitting documents into retrievable pieces.
• Citation — a pointer to the source passage of an answer.
• Grounded answer — an answer backed by retrieved content.
• Hallucination — an answer not backed by retrieved content.
• Context window — how much text the model can read at once.
• Prompt — the instruction/text sent to the model.
• System prompt — fixed instructions shaping model behaviour.
• Token (LLM) — a unit of model input/output (word or sub-word).
• Temperature — a setting controlling answer randomness.
• Natural language search — queries written as sentences.
• Query understanding — interpreting intent (entity, attribute, constraint).
• Intent detection — classifying what the user wants.
• Entity extraction — pulling named things out of a query.
• Attribute extraction — pulling constraints (size, color) out of a query.
• Semantic reranking — reranking with a semantic model.
• Multilingual search — matching across languages.
• Cross-lingual retrieval — matching a query in one language to content in another.

7. Merchandising and personalizationCopy Anchor Link

• Merchandising — manually steering what shows and in what order.
• Curation — human overrides on top of algorithmic ranking.
• Pinned result — a document forced to a fixed position.
• Boosted brand — raising products of a brand.
• Buried product — lowering a product's rank.
• Exclusion — removing a product from results entirely.
• Stock exclusion — hiding out-of-stock products.
• Rules — the declarative merchandising overrides.
• Campaign — a temporary merchandising setup (e.g. a sale).
• Recommendations — suggesting items beyond the explicit query.
• Bought together — co-purchase recommendations from order history.
• Frequently bought together — a common "bought together" widget.
• Related products — similar or complementary items.
• Alternatives — substitutes for a product.
• Personalization — adapting results to the individual user.
• User profile — the accumulated signals about one user.
• Behavioural signal — clicks, carts, purchases used for personalization.
• Cold start — the problem of personalizing with no user history.
• Popularity ranking — ranking by overall popularity for cold start.
• Trending — items rising in popularity recently.
• Query graph — relations between queries (synonyms, rewrites, related).
• Product graph — relations between products (variants, substitutes).
• Knowledge graph — an entity graph used for grounding.
• Journey — modelling a user's path across queries and sessions.
• Session personalization — personalizing within a single session.

8. E-commerce and catalogCopy Anchor Link

• Catalog — the full set of products.
• Product — one sellable item in the catalog.
• SKU — stock keeping unit; a unique sellable identifier.
• Variant — a specific configuration of a product (size, colour).
• Parent product — the umbrella item grouping its variants.
• Attribute — a product property (colour, size, brand).
• Category — a grouping of products in a hierarchy.
• Category tree — the hierarchical category structure.
• Product feed — a structured export of the catalog (XML/CSV/JSON).
• Price — the sell price of a product.
• Currency — the money unit (USD, EUR, RUB).
• Stock / inventory — how many units are available.
• Availability — whether a product can be bought.
• Out of stock — a product with no units left.
• Backorder — a product orderable before it is in stock.
• Product title — the name field used in results.
• Description — the long text describing a product.
• Image — the product picture shown in results.
• Thumbnail — a small product image.
• Product detail page (PDP) — the page for one product.
• Listing page — a category/search results page.
• Cart — the list of products a user intends to buy.
• Checkout — the final purchase step.
• Upsell — suggesting a higher-value product.
• Cross-sell — suggesting complementary products.
• Promotion — a discount or sale.
• Sale price — a temporary discounted price.
• Compare price — the original price shown struck through.
• Review — a customer rating and comment.
• Rating — the numeric score from reviews.
• Searchandising — merchandising applied to search results.
• Zero-results recovery — suggesting alternatives when nothing matches.
• Did you mean — a correction suggestion on zero results.
• Popular products — the best-sellers of an index.
• New arrivals — recently added products.
• Markdown — a price reduction.
• Margin — the profit per product.
• GTIN / EAN / UPC — barcode identifiers for products.
• Brand — the manufacturer or label.
• Manufacturer part number (MPN) — a manufacturer identifier.

9. Data and integrationsCopy Anchor Link

• Ingestion — bringing documents into the index.
• Connector* — a saved source (sitemap, URL, JSON, CSV) pulling documents.
• Full sync — re-importing the entire source.
• Delta sync — importing only what changed.
• Incremental update — applying only changes to the index.
• Reindexing — rebuilding a collection after a schema change.
• Data mapping — deciding which source fields go to which index fields.
• Field mapping — one source field → one index field.
• Transform — converting a value during ingestion (format, type).
• Feed — a machine-readable data export.
• Sitemap — an XML list of URLs for crawling.
• Crawler — a program that fetches pages.
• Webhook — an HTTP callback notifying of an event.
• ETL — extract, transform, load.
• Batch — processing many records at once.
• Rate limit — a cap on requests per time window.
• Idempotency — an operation that is safe to repeat.
• Idempotency key — a key preventing duplicate side effects.
• Retry — repeating a failed request.
• Backoff — increasing delay between retries.
• Queue — a line of pending jobs.
• Job — a scheduled or async unit of work.
• Schedule — the interval on which a connector runs.
• Log — a recorded event or message.
• Export — extracting data out of the system.
• Import — loading data into the system.
• Backfill — importing historical data.
• Schema change — modifying the field structure.
• Data validation — checking data before ingest.
• Enrichment — adding derived data to a record.

10. Platform, SaaS and multi-tenancyCopy Anchor Link

• Tenant* — one isolated customer workspace (organization, indexes, keys).
• Multi-tenancy — many tenants sharing one platform with hard isolation.
• Organization — the account grouping a customer's indexes and users.
• Workspace — the customer's working area.
• Index isolation — one tenant cannot see another tenant's data.
• SaaS — software as a service; delivered over the web.
• Cloud — running on provider infrastructure.
• On-premises — running on the customer's own infrastructure.
• Region — the geographic location of the servers.
• Deployment — the process of releasing a version.
• Rollout — gradually enabling a change.
• Rollback — reverting a change.
• Downtime — time the service is unavailable.
• Uptime — the share of time the service is available.
• SLA (service level agreement) — the promised availability/performance.
• SLO (service level objective) — the internal target behind an SLA.
• Quota — a limit on usage (requests, documents, indexes).
• Billing — charging customers for usage.
• Subscription — a recurring plan.
• Plan / tier — a level of features and limits.
• Usage-based pricing — charging by measured usage.
• Overages — usage beyond the plan limit.
• Seat — a licensed user slot.
• Provisioning — setting up a new tenant/resource.
• Deprovisioning — removing a tenant/resource.
• Scaling — adding capacity as load grows.
• Auto-scaling — scaling automatically with load.
• High availability — redundancy so the service stays up.
• Failover — switching to a backup on failure.
• Backup — a stored copy for recovery.
• Restore — recovering data from a backup.
• Snapshot* — a point-in-time copy of an index.

11. Security and complianceCopy Anchor Link

• Authentication — verifying who a user is.
• Authorization — deciding what a user may do.
• ACL (access control) — per-document visibility rules at query time.
• RBAC (role-based access control) — permissions by role.
• Role — a set of permissions (admin, editor, viewer).
• Permission — a single allowed action.
• API key — the credential authenticating API requests.
• Secret — a sensitive value (key, token, password).
• Token — a short-lived credential.
• OAuth — a delegated authorization standard.
• OAuth flow — the sequence of steps in OAuth.
• SSO (single sign-on) — one login for multiple services.
• SAML — an SSO protocol.
• Encryption — scrambling data so only authorized parties can read it.
• Encryption at rest — encrypting stored data.
• Encryption in transit (TLS) — encrypting data over the network.
• HTTPS — HTTP over TLS.
• Certificate — the credential proving a server's identity.
• Data residency — where data is physically stored.
• PII (personally identifiable information) — data identifying a person.
• GDPR — EU data-protection regulation.
• DPA (data processing agreement) — a contract governing data handling.
• Consent — user permission for data processing.
• Right to erasure — the GDPR right to be deleted.
• Audit log — a record of who did what.
• Audit trail — the chain of recorded actions.
• Compliance — meeting legal/industry requirements.
• SOC 2 — a security/controls audit standard.
• Pentest — a security test simulating an attack.
• Vulnerability — a security weakness.
• Threat model — an analysis of possible attacks.
• Least privilege — giving only the minimum required access.
• Private field* — a field only visible to authorized roles.

12. Developer and APICopy Anchor Link

• API (application programming interface) — a programmatic interface.
• REST — a common API style over HTTP.
• Endpoint — one URL path of the API.
• Method — the HTTP verb (GET, POST, PUT, PATCH, DELETE).
• Request — the client's call to the API.
• Response — the server's answer.
• Status code — the numeric result (200 OK, 404 Not Found).
• Header — a key-value in the request/response metadata.
• Body — the payload of a request or response.
• JSON — the common data format for API payloads.
• JSON field — one key-value inside JSON.
• Schema (API) — the declared shape of an API payload.
• OpenAPI — a standard for describing REST APIs.
• SDK — a client library (Node, Go, PHP, Python) wrapping the API.
• Client library — another name for an SDK.
• Package — a distributable unit of code.
• Version — a numbered release of the API or SDK.
• Semantic versioning (SemVer) — major.minor.patch.
• Breaking change — a change incompatible with older clients.
• Deprecation — marking a feature for future removal.
• Pagination (API) — paging through results.
• Cursor — a pointer for paginating a result set.
• Offset — a position from which to start a page.
• Filter (API) — a parameter narrowing results.
• Query parameter — a key-value in the URL.
• Rate limit (API) — a cap on requests per window.
• Rate limit header — metadata showing the limit and remaining.
• Retry-after — a header saying when to retry.
• Error response — a structured failure payload.
• Webhook (API) — an HTTP callback to your server on an event.
• Signature — a cryptographic check that a webhook is authentic.
• Embed widget* — a drop-in search UI added with a script tag.
• Script tag — an HTML <script> that loads the widget.
• Public key / token — a credential safe for client-side use.
• Secret key — a credential for server-side use only.
• Environment variable — a config value set outside the code.
• mCP (model context protocol)* — the protocol for exposing tools to AI agents.

---

The dictionary is deliberately grouped and kept to terms customers actually meet. To grow it, add a term under the matching section rather than creating a flat list — that keeps every definition near its context.