🔎 Ranking Algorithms

Ranking Algorithms#

Methods that order items by relevance or predicted value.

Important

✨ AI-generated content. This page was written with the assistance of an AI language model and is provided as a learning aid. Despite careful review, it may still contain mistakes, omissions, or out-of-date information. Whether you are new to the topic, a team lead, or a senior practitioner, treat it as a starting point rather than an authoritative reference: read it critically and independently verify anything you act on (code, commands, figures, and factual claims) against official documentation and primary sources before relying on it.

What it is#

A ranking algorithm orders a set of items — documents, products, ads, songs — so the most relevant or useful appear at the top. Given a query (a search term, a user profile), it scores items by predicted relevance, utility, or likelihood of interaction (click, purchase, watch) and sorts by that score. Search engines, recommenders, ad systems and social feeds are all ranking problems, and better ranking translates directly into user satisfaction and revenue.

Classical IR#

The oldest rankers are lexical. TF-IDF weights a term by how often it appears in a document against how rare it is across the corpus; BM25 refines this with term-frequency saturation and document-length normalisation, and remains a strong search baseline.

Learning to rank#

Learning to rank (LTR) trains a model from labelled relevance data, in three paradigms by what the loss looks at: pointwise scores each item alone (regression or classification — predict a click probability with logistic regression or gradient-boosted trees); pairwise learns from comparisons (“is A better than B?”, e.g. RankNet); and listwise optimises the whole ordering at once (LambdaMART, ListNet), often against a ranking metric like NDCG.

Neural rankers#

The newest models use embeddings (Word2Vec, BERT, Transformers) to capture semantic match rather than exact words — DSSM projects queries and documents into a shared space, and BERT-based rankers (monoBERT, ColBERT) read query and document in context, sharply improving relevance at higher compute cost.

How ranking is judged#

Ranking has its own metrics: NDCG (rewards relevant items near the top), MAP (average precision across queries), MRR (rank of the first relevant item), CTR, and precision@k / recall@k. The hard parts are personalisation (every user’s “best” differs), position bias (higher slots get clicks regardless of quality), scale (rank billions of items fast), and fairness to minority items.

Theme: Ranking & Interleaving · All terminology

Hint

Mind map — connected ideas

NDCG (Normalized Discounted Cumulative Gain) · Mean Average Precision (MAP) · Probabilistic Interleaving · Team Draft Interleaving (TDI) · Embedding · Online Experimentation Platforms

Hint

More in Ranking & Interleaving

Balanced Interleaving · DCG (Discounted Cumulative Gain) · Interleaving Tests · Mean Average Precision (MAP) · NDCG (Normalized Discounted Cumulative Gain) · Probabilistic Interleaving · Team Draft Interleaving (TDI) · TREC (Text REtrieval Conference)