Google patent about ranking from backlinks
What is really the value of a backlink? It depends on many factors that have been fully described in the patent 7716225 granted 11 May 2010 and entitled: Ranking documents based on user behavior and/or feature data.
It shows how a value is assigned to a link to another page based on the presentation and behavior expected of the user or user class.
The probability of clicking on a link is deduced from several factors, presentation, position, relevance, objectivity.
The presentation of a link
All factors which may influence the probability of the visitor to click on a link are taken into account and reinforces the value of this link.
- The font size and color.
If it is larger and more visible, it is more likely to be clicked on.
- The position of the link in the page.
Links placed early are more likely to be followed.
- The position in a list.
The links at the top of a list are more likely to be followed.
- The number of words in the text.
- The keywords in the text.
It is a relevant factor, both in the source page and to the target page.
- The fact that the anchor be commercial type.
If it contains words like "best", "first", etc. ...
A commercial link has less credibility.
- The words surrounding a link are taken into account.
They are intended to complement the text and interpret the relevance with respect to the target but also the relevance between the wording and content of the page (natural link).
- The fact that the link is text or in an image.
In the second case, the image is used to assess the importance of the link.
- The image size is taken into account.
A larger gives more weight to a link.
The independence between the source and target
A link on a site belonging to the same webmaster may be devalued. For this, besides the data known of the engine, it takes into account several factors.
- The fact that a link is on the same site.
An internal link is worth less than external link (even if it has the advantage its anchor associates keywords to the target).
- Or that the source and target have the same host.
It is assumed that this factor is weighted by the importance of the host.
The page with the link and the target
- PageRank is of course taken into account.
This is the first factor.
- The URL is taken into account.
If there is a relationship between keywords and those of the link, it reinforces the link.
A trust factor is given to each site.
- Sites associated with the target.
They are taken into account according to the patent.
- The number of links in the page.
The PageRank algorithm distributes the weight transmitted to the targets by sharing weight of the page between the links.
- Other keywords in the page.
This can reduce the relevance of a link.
- Other keywords in the titles.
- The topic of the page and of the target.
If the link is relevant, it is more likely to be followed. It is also the basis of contextual ads from Adsense.
- The topical cluster.
It is a ranking factor.
- If the link is on a redirected page.
A site having a link to a page on the same site through another site and a redirect, this page will be penalized. The link has a negative effect.
Links are often created in pages that are never modified by the webmaster. If the linked page is it changed, the link considered like a vote, must it retain its value? It gave a vote of something else actually. That's why patent for scoring of document takes into account the evolution in time, and the fact that new links are continually added.
It can be directly collected in various ways, on the results pages of the search engine, through the Google toolbar.
- It takes into account the language of the user.
The relevance is greater if the user's language is that of the target.
- The center of interest if known.
It can be inferred from queries leading to the source page.
- The number of users who choose a link.
- The total number of links clicked on the page.
- The change in behavior.
The evolution of how users click or not is taken into account.
The analysis of all these factors are used by Google to construct models that infer user behavior and hence the value to give to a link, and thus the ranking in the results of the linked page.
See also: Backlinks and ranking.