SEO link velocity: how Google reads the history of your links over time

Link velocity is the rate at which new links to a page appear over time. Patent US 8,407,231 B2, "Document scoring based on link-based criteria," claims a method that classifies a page as stale and lowers its score when "there has been a decrease in a rate or quantity of appearances of new links that point to the document over time." The patent was granted on March 26, 2013, from a family whose first filing, a provisional application, dates from September 30, 2003.

What does patent US 8,407,231 describe?

Patent US 8,407,231 describes a scoring method built on the time-varying behavior of the links that point to a page. It belongs to the same family as patent US 7,346,839, "Information retrieval based on historical data," the patent behind the domain age myth: it is a continuation of a 2006 application (now abandoned), itself a division of the 2003 application that became US 7,346,839. Its description is the one of that parent patent: it covers many kinds of history data (inception date, content changes, queries, traffic, domains, rankings), but its claims only cover links.

Its 8 inventors are Anurag Acharya, Matt Cutts, Jeffrey Dean, Paul Haahr, Monika Henzinger, Steve Lawrence, Karl Pfleger and Simon Tong. The patent holds 20 claims, including 3 independent ones (a method, a system and a storage medium), and 4 drawing sheets.

What exactly do the claims protect?

The claims protect a precise test on the appearance dates of a page's links. Claim 1 works in 4 steps:

1. Collect the history data: an appearance date for each link to the document.
2. Detect the decline: compare the oldest appearance date of all the links with the oldest appearance date of the group of newest links, the links whose dates fall "within a percentage of most recent appearance dates."
3. Classify the document as stale and decrease its initial score.
4. Rank the document against at least one other document based on the altered score.

The dependent claims add the details:

• The appearance date of a link can be "a date that a search engine web crawler first crawls a document that includes the particular link" (claim 3).
• The decline can also be measured by comparing the new links of "a pre-defined number of most-recent days" with the total quantity of new links (claims 4 and 5).
• The reverse case: an increase in the rate or quantity of new links classifies a document as fresh and increases its score (claim 6).
• Disappearing links: a document can also be classified as stale from the disappearance dates of its links (claim 7).

How does Google measure the freshness of a page's links?

Google measures link freshness by comparing recent links with older ones. The patent describes 4 measures:

1. New links over time: the number of new links "in the last n days compared to the number of new links since the document was first found."
2. The age of the most recent links: "the oldest age of the most recent y % of links compared to the age of the first link."
3. The age distribution of the links: the patent assumes that "the age distribution of a stale document will be very different from the age distribution of a fresh document."
4. Disappearing links: the date at which links disappear and the number of links that disappear in a given window of time.

The patent's example uses y = 10 and 2 sites first found 100 days ago. For site A, 10% of the links were found less than 10 days ago; for site B, none were. The metric gives 0.1 for site A and 0 for site B, and "may be scaled appropriately." A more detailed analysis of the distribution of link dates is possible as well: models can predict whether a distribution signals a site that is no longer updated, or one growing or declining in popularity.

"A downward trend in the number or rate of new links" signals a stale document, and "an upward trend may signal a 'fresh' document." The patent adds that a fresh document "might be considered more relevant, depending on the particular situation and implementation."

Why does a page's age change the value of its new links?

A page's age changes the value of its new links because the patent reasons in rates, not in totals. It can compute the rate at which links to a document are created, "as an average per unit time based on the number of links created since the inception date," and give "more weight to documents to which links are generated more often."

The patent's example: a document with an inception date of yesterday and 10 back links may be scored higher than a document with an inception date of 10 years ago and 100 back links, "because the rate of link growth for the former is relatively higher than the latter." The same paragraph sets the limit: a spiky rate of growth "may also signal an attempt to spam," in which case the score can be lowered.

The patent also gives a formula to adjust a link score by age: H = L / log(F + 2), where L is the link score from any link scoring technique (such as PageRank), F is the elapsed time since the inception date and H is the history-adjusted link score.

How are individual links weighted over time?

Individual links are weighted by a function that increases with their freshness. The freshness of a link can come from the date the link appeared or changed, the date its anchor text appeared or changed, or the date the page containing it appeared or changed. The patent notes that the last one "may be a better indicator of the freshness of the link based on the theory that a good link may go unchanged when a document gets updated if it is still relevant and good." To avoid refreshing every link after a tiny edit, each updated page is "tested for significant changes."

The patent adds other ways to weight links:

1. Trust: "government documents can be given high trust."
2. Authority: authoritative pages, determined in a manner similar to the original authority patent.
3. Freshness of the source: a page updated frequently (the Yahoo home page is the example) that suddenly drops a link to a document.
4. Stability: pages that have "a different featured link each day" are weighted lower than pages that are consistently updated and "consistently link to a given target document."
5. Staleness of the source: once a page is determined to be stale, "the links contained in that document may be discounted or ignored."

The score of a page can then rise or fall "as a function of the sum of the weights of the links pointing to it," and the technique can be applied recursively. A 2-year-old document can still be considered fresh if a share of its links are fresh, or if the pages that link to it are themselves fresh.

What does a spike of links mean?

A spike of links means either a topical phenomenon or an attempt to spam. The patent places this analysis in the detection of owners or their colleagues who "create links to their own document for the purpose of boosting the score." It states that "a typical, 'legitimate' document attracts back links slowly." A large spike can signal real news: "the CDC web site may develop many links quickly after an outbreak, such as SARS." It can signal spam as well, through "exchanging links, purchasing links, or gaining links from documents without editorial discretion on making links."

The patent names 3 sources of links without editorial discretion: "guest books, referrer logs, and 'free for all' pages that let anyone add a link to a document."

To tell the 2 apart, Google can consider mentions in "news articles, discussion groups, etc. on the theory that spam documents will not be mentioned, for example, in the news." It can also make exceptions for authoritative documents, "such as government documents, web directories (e.g., Yahoo), and documents that have shown a relatively steady and high rank over time." For such a document, an unusual spike of links is not treated as spam, and the system can allow "a relatively high or even no threshold" for the growth of its rank.

The same logic applies to rankings: a page that jumps in rankings across many queries may be topical or spam. Against spam, the patent describes "employing hysteresis to allow a rank to grow at a certain rate," or a maximum threshold of rank growth over a predefined window of time.

How does anchor text change over time?

Anchor text changes when a page or a domain changes its focus. The patent describes an expired domain bought by a new owner: old anchor texts keep pointing to it, so "the domain may show up in search results for queries that are no longer on topic." The fix is to estimate "the date that a domain changed its focus," from the date when the text of the document or its anchor text changes significantly, and to ignore "all links and/or anchor text prior to that date." In the same spirit, a significant change in the topics of a document "may indicate that the document has changed owners," so previous indicators such as score and anchor text "are no longer reliable."

The growth of anchor words matters as well. "Naturally developed web graphs typically involve independent decisions," while synthetic graphs "are based on coordinated decisions," which makes the growth of anchor words "relatively spiky." The patent cites 2 patterns: "a large number of identical anchors from many documents," or "deliberately different anchors from a lot of documents." Google can "cap the impact of suspect anchors" or use a continuous scale for the likelihood of synthetic generation to derive a multiplicative factor for the page's score.

What are independent peers?

Independent peers are unrelated documents linked to a page, incoming or outgoing. "A sudden growth in the number of apparently independent peers, incoming and/or outgoing, with a large number of links to individual documents may indicate a potentially synthetic web graph, which is an indicator of an attempt to spam." The signal is stronger when the growth corresponds to anchor text that is "unusually coherent or discordant." The impact of these links can then be demoted either as a binary decision ("demote the score by a fixed amount") or by a multiplicative factor.

The same distinction between independent and related links runs through the Panda patent, which counts only links from unrelated sites.

What does link velocity change for your SEO?

Link velocity means that the history of your links matters as much as their number. 5 consequences follow:

1. Keep earning new links. A falling rate of new links can mark a page as stale, and its score can then be decreased.
2. Grow at a natural pace. Legitimate pages attract links slowly; a sudden spike without news to explain it looks like bought or exchanged links.
3. Pair link peaks with real news. Mentions in news articles and discussion groups are the patent's own way to tell a topical phenomenon from spam.
4. Avoid links without editorial discretion. Guest books, referrer logs and pages where anyone can add a link are the patent's own examples.
5. Check the history before buying an expired domain. Links and anchors from before a change of focus can be ignored, so an expired domain may not carry its old authority to a new topic.

The patent describes what Google's system can do. It does not confirm which link history signals Google uses today, or their weight.