Results Ranking
From Yfittu
- Audience
- This is intended for all developers interested in what lies behind the scenes or anyone that is just curious. Not intended for users.
- WARNING
- This page is only a draft now. It gives some concept but a lot of rewording and reorganisation is necessary. (Of course, this paragraph should be removed when this page would be clear and well organised)
- Object of this page
- describe how pages are ranked to sort results by relevance. The object of this page is not the ranking for indexation designed to provide a good distribution of the indexing job and a more important place for more important pages/sites.
[edit]
General site and page relevance scoring
Note: this scoring is also taken into consideration for calculating the score for indexing pages or not (see the indexing robot).
Criteria:
- document analysis (antispamdexing, antiphishing, global lexical, spelling evaluation) : that concerns what we could assume just by analysing the page, without taking other pages into consideration
- number of links FROM other sites that points here and where they come from (that should not in any case penalize: we can't accept that a site could have a negative impact on another by using spamdexing technics)
- number of internal links (to detect the most important pages on a site)
- links TO other sites (here links to some sites may implies penalties: a site which attempts to use spamdexing technics should be penalized)
- date of last update of the page known by the indexing robot
- and a lot of others ...
The evaluation of each criterion (by a single peer) is made at indexing time, not at search time, so a good analysis is preferred, even if it takes some CPU time.
[edit]
Relevance according to match with search terms
Criteria:
- best distance between the words
- number of times they occurred
- importance of the searched text in the page (in absolute and as a rate, depends on both bold, title attributes, position: beginning / end or middle ...)
- title of links that points to this page
- text in the url
Evaluation on a search request but all what can be done at indexation time in order to speed up this evaluation should be done.
