If you influence behavioral factors in a natural or artificial way, then this note will help you understand whether you are moving in the right direction.

This time there is almost no gag - the list of synthetic behavioral factors was published at the goodwill of Yandex in the report Through-the-Looking Glass: Utilizing Rich Post-Search Trail Statistics for Web Search.

Yandex itself dates the article to September 1, 2013. Was presented at the CIKM 2013 conference, held from October 27 to November 1, 2013 in Burlingame (USA)

There is a mention of the report on the Yandex website, but the text itself is still not there. Do you realize that you've said too much? 🙂

NB: I express my gratitude to Ilya Zyabrev AlterTrader Research ltd for providing the text of the report.

Yandex behavioral ranking factors

QueryDomCTR– average CTR value of all domain documents for a given query

QueryUrlCTR– the average CTR value of a specific document for a given request.

QDwellTimeDev– standard deviation (deviation) from the average time spent on the document upon request. Can be used to filter out PF cheats.

QDwellTime– this parameter is not mentioned in the report, but it is obviously used as a ranking factor. Because if we calculate the standard deviation for a random variable, then we must also know the expected value (also known as the average value) of this variable. Accordingly, this is the average time a visitor spends on a document upon request.

AvSatSteps– the average number of satisfied steps on the site. A satisfied step is following an internal link after 30 seconds of being on the document. It is important that the average value of such steps is only ~0.2 or less per domain, regardless of the subject of the site.

NB: As follows from this report, Yandex “knows” what topic the site page belongs to. Based on our own set of second-level domains with manually defined topics (as I understand it, this is nothing more than Ya.Catalog, possibly expanded by Dmoz.org) and using a naive Bayes classifier, any document from the index is assigned to a particular topic.

AvDwellTime– the total average time a visitor spends on a document for various search queries.

DwellTimeDev– standard deviation (deviation) of the time spent on the site. It can also be used to track PF markups.

90thDwellTime– this is the top decile, also known as the 90th percentile of average time spent on the site. Allows you to discard added AvDwellTime and QDwellTime.

10thDwellTime– This is the bottom decile of average time spent on the site. Allows you to define doorways. It is obvious that Yandex expects improvements from “white” webmasters here.

TimeOnDomain– total time spent on the site. For all requests for any documents.

CumulativeDev– standard deviation (deviation) from the average time spent on the site

A few snide remarks

1. Are you afraid that boosting PF reduces conversion and this negatively affects rankings? Don’t anger the SEO god - Yandex can only count satisfied steps. He doesn't even think about conversion. .
2. To satisfy Yandex engineers, you should break large articles into small ones. Very small. Because the average person reads less than 300 words per minute. And engineers are interested in a click after 30 seconds. By this point, you've already read about 300 words in this short piece. By comparison, a good review article is considered to be at least 1,000 words long.
3. For the same reason, you should not place either internal or especially external links at the beginning of the document. So that there are no unsatisfied transitions.

Instead of a conclusion

This Yandex report perfectly explains why boosting behavioral ranking factors for 3-4 queries does not work. Namely, this is the number of queries that a regular (median) optimizer tries to generate. The average is only 9. A successful optimizer generates an average of 40 queries and more. This is the only way to significantly influence, for example, AvDwellTime and TimeOnDomain.

I presented this data in June 2013 on

March 18, 2013 at 2:04 pm

Behavioral Ranking Factors

• SeoPult company blog

Until a computer starts thinking like a human, it won't be able to tell a bad site from a good one... the way a human would. In fact, search engines have in their arsenals techniques for collecting and analyzing data, with the help of which silicon brains can easily put meat experts in the belt.

Let’s make a reservation right away - by “good” site we mean “worthy of taking a place in the search results for a specific key query”; we will not plunge into the jungle of site-building aesthetics.

So, without going into details, search engines now comprehensively use three approaches: ranking pages by authority (for example, the PageRank algorithm that brought Google popularity), behavioral factors (analysis of the actions of real visitors to real sites), and machine learning (for example, Matrixnet) Yandex, which trains algorithms by assessing samples by expert assessors, essentially links and balances the first two approaches).

Ranking by authority worked very well in the early stages of the Internet, but later the “too mathematical” nature of this approach allowed optimizers to use tricks that exploit system weaknesses found during experiments. The quality of search results suffered, search engines introduced amendments, additional formulas and coefficients, filters and sanctions, but a truly major breakthrough was made when it became possible to rank sites based on the preferences of their real live visitors. Analysis of behavioral factors is more objective than any personal bias (both expert and profane), since it works with the preferences of a large sample of the target audience.

How search engines collect data

1 Statistics systems(Google Analytics and Yandex.Metrica). Almost all website owners want to have information about traffic and all audience actions. The best, and even free, tools for this are provided by search engines, but in return they receive huge amounts of data.

2 Browsers. I Internet Explorer runs on Bing, Chrome runs on Google, and Yandex also has its own product in this niche. Although Chrome, for example, deep, deep in the settings, where only hacker-heads will climb, provides the ability to uncheck the checkbox “Automatically send usage statistics and crash reports to Google,” we cannot guarantee that this deprives the good corporation of access to the data it needs . In general, the flow of data from branded browsers is quite significant; it allows you to additionally cover the segment of sites without statistics systems (or, more often, with competitors’ statistics systems).

3 Browser add-ons. You can assess the need of search engines for traffic data by aggressively promoting Yandex.Bar. Turning any browser into a “branded” one, the add-on diligently sends traffic statistics to its native data center.

With the above products, search engines receive almost complete information about the behavior of the audience of each indexed site. The next logical step is to show higher those sites that, other things being equal, evoke a more positive reaction from visitors. This, of course, has its own subtleties: in some topics, the time spent viewing a page will be the main factor in a positive rating, in others (for example, if the user needs to take one look at the page to perform the required action) it does not play a special role. In some places, browsing depth is very important, but if the site consists of one page, then sometimes it is not worth denying it high positions. This is where data interpretation and segmentation, as well as machine learning, come into play (if assessors consistently rate high-quality one-page promotional sites highly, the search engine will learn to exclude browsing depth from the list of important behavioral factors for similar resources).

Key Behavioral Ranking Factors
1 Bounce Rate(bounce rate) - the percentage of visitors who left the site after viewing the login page. For sites that require several transitions to other pages - and these are the majority - this is a very good criterion for quality and relevance to the topic. The visitor leaves the site either because he found what he needed and did what he was going to do (and what the site owner required), or because he did not like the site or is not relevant to the search query. Try to reduce the bounce rate - increase relevance, improve design and UX, make landing pages more understandable, attractive, and so on. Of course, there will never be a 100% “assimilation” of the audience, but we must strive for this. And not least because of the consideration of behavioral factors by search engines, but because of conversion - the bounce rate is directly related to the site’s ability to turn “visitors” into “buyers”.

2 Time spent on the site. A good quality criterion in the vast majority of cases. If a high rate is achieved not due to the incomprehensibility of the content and confusing navigation. You can increase viewing time using the simplest everyday logic: give visitors something that interests them, and they will devote their time to studying these materials. These could be articles, photo galleries, videos, some services such as mortgage calculators (in the subject of the site, of course), etc. All engagement techniques should not harm conversions, so don’t mechanically add everything to the page.

3 View depth. An important criterion for content projects. You can increase depth through thoughtful navigation and cross-referencing, and, of course, interesting content. Many sites try to increase depth by breaking large articles into several parts located on different pages, but this practice is justified only if visitors are highly motivated to read the entire article (this works well for reviews of computer components, but for the continuation of a “humanitarian” article about , how can we reorganize the Rabkrin, many will refuse to move).

4 Return to re-search. If a visitor returns from the site back to the search, it means he did not find what he was looking for. This parameter can be controlled only by increasing the relevance of landing pages to queries, as well as maintaining a competitive level of prices for goods and services.

5 Returning to the site is not from a search. If a visitor has bookmarked the site or remembered the address, this will be a significant plus in favor of the resource. However, you should not intrusively suggest adding a site to bookmarks; this must be done subtly and tastefully.

6 The nature of the mouse cursor movement and the pattern of movement around the site. Statistics systems collect data not only about where the visitor clicked, but also how he moved the cursor. This is necessary to build “heat maps of attention”, as well as to filter out attempts to cheat behavioral factors with scripts. Patterns of live visitors are quite difficult to emulate, which is why, by the way, many sites that tried to boost user factors in the first months of their implementation quickly fell down in search results or were banned - search engines noticed that the cursor was controlled not by real people, but by programs. Analysis of the heat map and recordings of viewing sessions will allow, with enough time and meticulousness, to identify and eliminate obstacles on conversion routes.

7 Snippet click-through rate (CTR). The more people click on your snippet (a short description with a link in the search results), the better the search engine treats you. This is logical: if the snippet is relevant to the query and attracts the attention of users, then the site is likely to provide a high-quality response to the query. There are ways to control the snippet, and this is worth paying attention to - quick links, the right title, good text will help increase both traffic and positions.

8 Social media buttons. If the installed buttons (it is best to install not scripts like AddThis, but native buttons of the social networks themselves) are clicked, this not only increases the number of subscribers to your pages in these social networks, but is also an important signal of the quality of the site for search engines. Install the buttons as early as possible - each subscriber will be a significant plus.

conclusions
Search engines analyze data on behavioral ranking factors quite objectively and qualitatively. Do not try to manipulate them directly (using scripts, buying untargeted traffic, etc.): this will lead to sanctions and will not bring any benefit. It is much more effective and important to actually increase the quality of the site, its attractiveness and conversion. Then users will behave the way you want, and all of the above indicators and, accordingly, the site’s position will grow.

Behavioral factors (PF) - These are characteristics of a page or site that describe the behavior of the average visitor and influence search engine rankings.

Significance for promotion

Behavioral factors are extremely important for Yandex - if a site has poor user metrics, it has little chance of ranking high in search results for competitive queries. Along with tracking the actions of real users, Yandex attaches great importance to the opinions of specialists who analyze the site according to a set of parameters and evaluate the overall impression. For Google, the PF is much less important, but it is known for sure that it takes into account some factors when assessing the relevance of a page to a query.

Collecting information about user behavior

Search engines have several tools to collect information about user behavior on a website.

Web analytics systems

The web analytics system counters installed on the site (Google Analytics and Yandex.Metrica) provide search engines with detailed information about user behavior. Of course, the PS cannot install the system code without the knowledge of the webmaster, but webmasters themselves willingly post this code, because web analytics tools are useful for them too.

Browser add-ons

Yandex is actively promoting its own brainchild - the add-on for the Yandex.Elements browser (until 2012 - Yandex.Bar). This is a small panel with quick access to various Yandex services and displays information about the weather, exchange rates, etc. It works in much the same way as the Metrica code: it collects information about the behavior of a site visitor who uses a browser.

Search engine browsers

Both Google and Yandex have their own browsers - Google Chrome and Yandex.Browser, respectively. Both are also used for data collection.

operating system

The most popular mobile operating system, Google Android, is owned by Google and comes with many additional programs and services that regularly access Google servers. It is likely that the system also sends information about user actions on websites.

Clicks on snippets and behavior on search results pages

The search engine collects behavioral information from . If users are more likely to click on lower SERPs, then those sites are more relevant to the query. If, having reached a certain site from the search results, users do not return to the search results, this is a signal that they have found the answer to their question.

Key Behavioral Factors

• the resource as a whole speaks of its popularity on the Internet.
• search engine. Here you should highlight the first and last clicks on the search results. It is generally accepted that for Yandex the most relevant pages are the first and last of the many pages opened from the search results.
• Time spent Online- one of the most important indicators of website quality. A good website will keep a visitor for a longer period of time, unlike a bad one, which will be abandoned almost immediately.
• - the number of pages viewed per visit is taken into account.
• . A refusal is a situation where a user views only one page and spends no more than 15 (according to other sources - 20) seconds on it. Such a short visit indicates that the resource is of poor quality and not as requested.
• Return rate. If users return to the site pages, it means they find something they need there.
• Direct transitions And transitions from social network pages can also be a sign of the resource's usefulness.

Working with Behavioral Factors

A webmaster can and should work to improve the site’s PF: internal optimization and work with PF today are the main methods of search engine promotion. The ways to improve user metrics are obvious: the site simply needs to be useful and convenient for visitors.

Hello everyone!

When ranking in search results, Yandex and any other search engine takes into account a lot of things, including behavioral factors (PF). Let's try to understand their impact on search engine promotion. Let's go!

Behavioral factors are indicators of how users behave on your web resource. And to be precise, then:

• How much time do they spend on the site?
• Which ;
• How many pages are viewed when visiting;
• How to interact with various elements of the site;
• Well, there are many other factors.

All this is taken into account by search engines and is subsequently used to rank sites in search results: the better the PF, the higher the site’s position. After all, Yandex and Google have recently focused specifically on the usefulness of search results for their users.

How do PSs obtain data about user behavior?

In order to transfer data about your visitors to Yandex, it’s quite simple. Have you ever wondered why it is free? The exact same situation happens in Google, only with .

Of course, PSs cannot rely solely on analytics systems, so there are several other ways by which they learn the behavioral factors of a site:

1. Browsers. Nowadays, you can forget about privacy, since each browser can transmit information about you to search engines. And browsers from Yandex and Google: Yandex.Browser and Chrome, respectively, transmit an even larger array of data about you;
2. Special extensions for browsers. Probably everyone has pre-installed add-ons from popular search engines, for example, an email client. They may also collect information about user behavior;
3. And there are many, many more ways to obtain data of interest to the PS.

Thus, it turns out that we are all godlessly burned out, which makes the life of an SEO optimizer only more difficult. After all, search engines put the user at the forefront, which means that all sites presented in the search results must be designed so that he receives the most complete answer to his question with convenience.

Promotion by behavioral factors

Let's think about how behavioral factors affect a site's ranking in search results. So:

The search engine has some data about each site, among which yours is presented, for a specific request. You are in 9th position, and your main competitor is in 3rd. Your task: to enter the TOP-3.

To complete your task, you hired a copywriter with a high level of texts and filled your website with content. And they did it in such a way that the user received the most detailed answer to his question. Traffic flows to the page for this specific request, but if you rank in the TOP 3, you will receive even more visits.

The main competitor’s page is inferior in quality to yours, and the user who goes to it does not receive a proper response and does not stay there for more than a minute and a half. Whereas visitors to your page stay on it for almost two minutes more, and even show activity in the form of clicks on various links. The search engine noticed this.

What do you think will happen next? If your page has more users and gives a more comprehensive answer, which makes it stay for a long time, compared to your main competitor, you will take a more advantageous position, and the competitor will be pushed down.

From the above example we can conclude: in order to have good positions in the PS, you need to make a website based on user needs. That is, the web resource must be convenient for interaction And useful. This is what the principle of promotion by behavioral factors is based on.

However, due to this mechanism of search engines, one insidious “black” has appeared - cheating of behavioral factors.

Cheating behavioral factors

This method was once very popular among SEO optimizers, as it was highly effective. But soon the PS saw this trend and put forward countermeasures to such pampering in the form of a filter for cheating behavioral factors.

The principle of operation of the method is as follows:

1. You register in the PF promotion service;
2. Choose a method: catch up with robots or real people;
3. Then you feel a systematic improvement in behavior;
4. PSKs also see this and rank you higher in search results.

At first glance, everything is simple, but the filter quickly calculates all this action and gives a harsh punishment. And he calculates it like this:

Robots, that is, automatic PF cheating, are characterized by similar IP addresses, similar browsers, similar devices, and similar behavior. By comparing all data about visitors, the filter identifies the crime and punishes it.

If you decide to catch up with real people, that is, use manual promotion, then the filter can calculate this from the history of interaction with the search for the person who visits your site. A normal user will not become interested in cars, culinary recipes and IT technologies or anything else in one day.

Therefore, experienced webmasters do not recommend resorting to such services, because it is much better to improve the behavioral factors of the site in natural ways.

How to improve behavioral factors?

There are several “white” ways to improve PF:

1. Competent;
2. Filling the site with useful, interesting and comprehensive material. Suitable for this:

Search engines have been successfully fighting spam in search results for a long time, constantly improving the ranking formula and supplementing it with new factors that are less noisy than, for example, the same reference. In this article, we will talk about behavioral metrics and discuss their impact on search results.

So what are behavioral factors? If we summarize all the information that is freely available, we can formulate the following definition:

Behavioral factors —This is the totality of all user actions on search engine results pages and the website.

We all remember that Yandex has always promoted the creation of sites for people that would satisfy the needs of users who switch to them. And behavioral factors are the best way to display this information in digital form.

The behavioral factors themselves can be divided into “click” and “host” factors. The former refers to those factors that the search engine tracks when the user interacts with the SERP. The second are those that can be gleaned from the statistics of the site itself.

List of behavioral factors

The list of synthetic behavioral factors was published by experts from Yandex in the report Through-the-Looking Glass: Utilizing Rich Post-Search Trail Statistics for Web Search. The following ranking factors have become known:

1. QueryDomCTR— the average CTR value of all documents in the domain for a given request.


2. QueryUrlCTR— the average CTR value of a specific document for a given request.


3. QDwellTimeDev— standard deviation (deviation) from the average time spent on the document upon request.


4. QDwellTime— the average time a visitor spends on a document upon request.


5. AvSatStelis— the average number of satisfied steps on the site. A satisfied step is following an internal link after 30 seconds of being on the document.


6. AvDwellTime— the total average time a visitor spends on a document for various search queries.


7. DwellTimeDev— standard deviation (deviation) of the time spent on the site.


8. 90thDwellTime— this is the top decile, also known as the 90th percentile of average time spent on the site.


9. 10thDwellTime is the bottom decile of average time spent on the site.


10. TimeOnDomain— total time spent on the site. For all requests for any documents.


11. CumulativeDev— standard deviation (deviation) from the average time spent on the site.

Of course, it is worth making a reservation that these are not factors from the Yandex ranking formula (and this is important to understand), but only factors that were used on a certain test sample of documents as part of the ongoing research. However, it can be assumed that the ranking algorithm takes into account similar metrics.

In my opinion, the most interesting for further study are QueryDomCTR, QueryUrlCTR, QDwellTime, AvSatSteps, AvDwellTime, TimeOnDomain. Because they are the easiest to both understand and use.

The six factors identified can be divided into the following groups:

Group of “click” factors. As we can see, this includes metrics related to the clickability of the entire site or its individual pages in search results.

In my opinion, it suggests itself: increasing clicks on individual queries (QueryUrlCTR metric) may be ineffective if the overall click-through rate of the resource is low (QueryDomCTR metric). Agree that if a site has only 5 requests out of 100, this is strange, and Yandex, apparently, knows how to successfully deal with this.

What to do?

Work on the site description on search results pages. Make your descriptions better than your competitors. You can read how to achieve this in one of our previous articles.

Group of “host” factors. This includes factors related to the time the user spends on the site. It is worth noting the fact that it depends on the subject of the resource and cannot be unified.

What conclusion can be drawn from these metrics?

• It is important to keep the visitor on the page to which he went from the search results. But you shouldn’t forcibly “handcuff him to the radiator.” Provide him with as much useful information as possible. For example, if you have an online store, then it wouldn’t hurt to place a video describing their characteristics on product cards.
• Where possible, encourage the user to move around the site: offer, for example, to visit pages on a topic of interest or look at products that may be of interest to him.
• Work with the entire resource as a whole, and not just with promoted pages, since Yandex can look at the total time users spend on the site for all requests, and not just for those promoted within the SA.

The importance of behavioral factors

It is worth touching on the importance of behavioral factors in current realities. For a long time, link factors ruled the roost, but now Yandex wants to reverse this trend. Perhaps not right away, but over time, I'm sure he will be able to do it. Already now, for some types of requests, the influence of behavioral factors is taken into account more than the same reference ones.

This is due to the fact that the latter are currently heavily “noisy” by optimizers who are trying to manipulate search results in this way. Naturally, this state of affairs does not suit the PS, and, in particular, Yandex is trying to level out “spam” factors, replacing them with others.

In conclusion, I would like to wish all readers good luck on their way to the TOP!