Is a posting a full copy of a document?

Usually it is an index record pointing to a document and optionally storing associated information, rather than the complete text.

Does every posting store positions?

No. Positions are an index-design choice needed for particular forms of phrase or proximity checking.

Can postings alone establish that a result is useful?

No. They provide matching evidence; usefulness depends on the query, document content, ranking, and task.

AI Glossary / Natural language and speech / Intermediate

Postings

Definition

Postings are the records in an inverted search index that associate an indexed term with documents containing it. A basic posting identifies a document. Richer postings can also hold the term's frequency or positions within that document. The postings list for a term groups these records so a search engine can find matching documents efficiently.

What it means

Here, postings means search-index records, not social media posts or job advertisements. An inverted index starts with a term and points to its occurrences across a collection. This reverses the everyday reading direction in which someone opens a document and encounters its words.

The dictionary of an index identifies the indexed terms; the postings provide their document associations. A document identifier is an internal reference, not necessarily a public web address. Whether a posting includes counts, positions, or other information depends on the index design and the searches it must support.

How it works

During indexing, text is divided into tokens and processed according to the system's rules. Those rules might normalize case or word forms. The system associates resulting terms with document identifiers and consolidates repeated occurrences in the same document. A basic list therefore need not repeat a document identifier once for every occurrence.

At query time, the engine obtains lists for the requested terms. An AND query can combine them by retaining document identifiers present in both. Sorted lists permit a coordinated traversal: matching identifiers are retained, while the smaller unmatched identifier is passed over. Other query types need different combinations or additional checks.

Positions make a further distinction possible. Knowing that two words occur somewhere in a document does not establish that they occur next to each other in the requested order. Positional postings let the engine test that relationship. The resulting evidence still concerns indexed text and tokenization rules; it does not establish that the document's claims are true or that the result answers the user's underlying question.

Three practical examples

1. Product-document search: A help system uses postings to identify manuals containing both a model name and an error term before considering how to rank the matches.

2. Phrase lookup: A document archive checks term positions to distinguish “red valve” from separate mentions of red equipment and a valve elsewhere.

3. Index diagnostics: A search engineer examines the postings for an unexpectedly missing term to determine whether the document was indexed and how its words were normalized.

Think of it this way

Think of postings as the page references under a heading in a back-of-book index. The heading resembles an indexed term, and the references indicate where to look. The analogy is limited: a search engine can store document-level frequencies and exact token positions, and its indexing rules need not match a human editor's choice of topics.

Worked example

Consider three original toy documents, with lowercase words and one-based token positions:

Document 1: “red valve red”
Document 2: “blue valve”
Document 3: “red pipe valve”

The positional postings for red are document 1 at positions [1,3] and document 3 at [1]. For valve they are document 1 at [2], document 2 at [2], and document 3 at [3]. Red has document frequency 2, although it occurs three times across the collection. Its term frequency within document 1 is 2.

The Boolean query red AND valve returns documents 1 and 3 because both terms occur in each. The exact phrase “red valve” returns only document 1: valve at position 2 immediately follows red at position 1. In document 3, pipe separates them. These lists and counts are independently checked teaching data, not measurements of a real search service. They show why document frequency, occurrence counts, and adjacency answer different questions.

Illustrative case study

Fictional case: Lantern Maintenance notices that staff searching for a component phrase receive documents containing its two words far apart. The search team inspects a small test collection and confirms that the current query checks only document membership.

It adds a separately tested phrase-search option based on stored positions and explains the distinction in the interface. Staff can still request broad term matches when appropriate. The case demonstrates how index evidence affects a search feature; it does not claim a particular speed improvement or assume phrase matching always produces the most useful result.

Common misconceptions and mistakes

Do not confuse a postings-list length with the number of times a term appears. In the basic one-record-per-document arrangement, list length counts documents containing the term. Repetition within a document is a different quantity.

Likewise, two terms sharing a document does not prove a phrase match. Stored positions and the system's phrase rules matter. A missing posting may reflect preprocessing, an indexing omission, or an outdated index; it is not by itself proof that the original document lacks the visible word.

Business use

For business search, postings are useful diagnostic evidence when results behave unexpectedly. Build small, known test collections containing repeated terms, separated terms, and exact phrases. Check that indexing and querying use compatible text processing, and compare retrieved documents with the intended search behavior before drawing conclusions about relevance.

Check your understanding

Red appears twice in document 1 and once in document 3. What are its document frequency and total occurrence count?

Answer: Its document frequency is 2 and its total occurrence count is 3. The repeated occurrence in document 1 does not create another distinct matching document.

Related terms

Inverted index
Tokenization
Positional index
Information retrieval
document frequency
term frequency
Boolean retrieval
phrase query

Learn next

Learn inverted indexes first, then work through intersections of small document-ID lists. Add positions and test exact phrases separately. Finally, explore how ranking uses matching evidence without treating term occurrence as a complete account of relevance.

Sources

Manning, Raghavan, and Schütze, Introduction to Information Retrieval, primary textbook sections consulted:

A first take at building an inverted index: https://nlp.stanford.edu/IR-book/html/htmledition/a-first-take-at-building-an-inverted-index-1.html

Processing Boolean queries: https://nlp.stanford.edu/IR-book/html/htmledition/processing-boolean-queries-1.html

Positional indexes: https://nlp.stanford.edu/IR-book/html/htmledition/positional-indexes-1.html

The three-document collection, worked results, and company case are original illustrations.

Previous: Posterior sampling Next: Postposition