relevance

information science A measure of how closely a given object

(file, web page, database record, etc.) matches a user's

search for information.

The relevance algorithms used in most large web searchengines today are based on fairly simple word-occurence

measurement: if the word "daffodil" occurs on a given page,

then that page is considered relevant to a query on the word

"daffodil"; and its relevance is quantised as a factor of the

number of times the word occurs in the page, on whether

"daffodil" occurs in title of the page or in its META

keywords, in the first N words of the page, in a heading,

and so on; and similarly for words that a stemmer says are

based on "daffodil".

More elaborate (and resource-expensive) relevance algorithms

may involve thesaurus (or synonym ring) lookup; e.g. it

might rank a document about narcissuses (but which may not

mention the word "daffodil" anywhere) as relevant to a query

on "daffodil", since narcissuses and daffodils are basically

the same thing. Ditto for queries on "jail" and "gaol", etc.

More elaborate forms of thesaurus lookup may involve

multilingual thesauri (e.g. knowing that documents in Japanese

which mention the Japanese word for "narcissus" are relevant

to your search on "narcissus"), or may involve thesauri (often

auto-generated) based not on equivalence of meaning, but on

word-proximity, such that "bulb" or "bloom" may be in the

thesaurus entry for "daffodil".

Word spamming essentially attempts to falsely increase a web

page's relevance to certain common searches.

See also subject index.

(1997-04-09)