Glossary

Plain-language definitions for the ecommerce search, product discovery, and AI relevance concepts we work with every day.

A

Autocomplete

Inline suggestions that complete a search query as the shopper types, surfacing products, categories, or popular searches before they hit Enter.

B

BM25

A probabilistic ranking function that scores how relevant a document is to a query based on term frequency, document length, and rarity across the corpus.

C

Click-Through Rate (CTR)

The percentage of search-result impressions that result in a click — the primary engagement signal for measuring search ranking quality.

Conversion Rate

The percentage of searches that result in a purchase — the dollar-value endpoint of search quality, downstream of CTR and add-to-cart.

Cosine Similarity

A distance metric that measures the angle between two vectors, ignoring their magnitude — the default similarity score for embedding-based search.

Cross-Encoder

A transformer that scores query and document together in a single pass — slower than bi-encoders but much more accurate, used as a re-ranker on the top-N candidates.

D

Did You Mean

A spell-correction suggestion shown when the user’s query produces few or zero results, offering a likely-intended alternative.

E

Embedding

A dense numerical vector that represents a piece of text, image, or other input in a learned space where geometric distance reflects semantic similarity.

Entity Recognition

The NLP step that finds and classifies named things in a query — brands, sizes, colors, materials, locations — so they can be routed to facets instead of full-text search.

F

Faceted Search

Filter-driven refinement that lets shoppers narrow results by structured attributes — size, color, brand, price band, rating — without writing a new query.

G

Generative Search

Search that returns a synthesized answer (often citing source products) generated by an LLM, rather than a ranked list of results.

H

Hybrid Search

Search that fuses lexical (BM25) and semantic (vector) retrieval, combining their scores so exact matches and meaning-based matches both surface.

I

Inverted Index

The data structure underneath classical search engines — a map from each term in the corpus to the list of documents containing it, enabling sub-millisecond keyword lookup.

K

Knowledge Graph

A structured representation of products, attributes, and the typed relationships between them — used to power facet hierarchies, related-product widgets, and disambiguation in search.

L

Learning to Rank

A family of supervised ML techniques that learn a ranking function from labeled query-result pairs, replacing hand-tuned scoring formulas with a model trained on relevance data.

Long-tail Query

A search query that appears rarely or only once in your search logs — individually low-traffic but collectively dominant, and historically the hardest queries to rank well.

M

Mean Reciprocal Rank (MRR)

A ranking metric that averages 1/(position of first relevant result) across queries — rewarding rankers that put the right answer first, indifferent to deeper ranking quality.

Merchandising

Manual or rule-based overrides that pin, boost, bury, or replace specific products in search results — the human layer on top of algorithmic ranking.

N

NDCG

Normalized Discounted Cumulative Gain — a ranking quality metric that rewards relevant results highly placed in the result list and discounts those buried below the fold.

O

Out-of-Stock Filtering

The decision of whether to hide, demote, or surface-with-warning products that are currently unavailable — a small policy with measurable revenue impact.

P

Personalization

Tailoring search results to the individual shopper based on their history, behavior, or declared preferences — a per-user re-ranking layer on top of relevance.

Precision

Of the results returned, what fraction are relevant — the “quality, not quantity” IR metric, complementary to recall.

Q

Query Understanding

The NLP pipeline that classifies, rewrites, and enriches a raw search query before retrieval — extracting intent, entities, attributes, and constraints.

R

Re-ranking

A second-stage scoring pass that takes the top-N results from a fast retriever (BM25 or vector) and re-orders them with a more expensive, higher-quality model.

Recall

Of all the relevant items in the catalog, what fraction did the search return — the coverage metric, complementary to precision.

S

Semantic Search

Vector-based retrieval that matches by meaning rather than exact keywords, using embeddings to find products even when query and title share no words.

Synonyms

Equivalent terms a search engine treats as interchangeable so a query for one returns results containing the other (couch ↔ sofa, sneaker ↔ trainer).

T

TF-IDF

Term Frequency × Inverse Document Frequency — the classical weighting scheme that scores a term high when it appears often in a document but rarely across the corpus.

V

Vector Search

Approximate nearest-neighbor (ANN) retrieval over high-dimensional embeddings — the engine behind semantic search, recommendation, and visual similarity.

Visual Search

Image-to-product search — upload or paste a photo, get matching catalog items based on visual similarity rather than text.

Z

Zero-Result Rate

The percentage of search queries that return no products. The single most actionable health metric for an ecommerce search engine.