Yandex's Atom platform is an Internet adapted for everyone

    Just at YaC 2013 Yandex announced the new Atom platform . She uses many technologies that were created for different tasks to solve one big one - to change the Internet so that each of us ceases to be an abstraction for him, but becomes a person with his own character and interests.

    image

    Gradually, everyone came to understand that the Internet is determined not by documents, but by people. It is connected with reality and consists of the needs, preferences and tasks of people - like a world of atoms. The Atom program is about a person; she puts forward and tests hypotheses about what he wants, what interests him, what he needs on a particular site.


    Every step in the development of our technologies was necessary in order to cope with the growth of content on the Internet and to be able to show people the answers that were right for him. In 2009, we launched the Arzamas search platform . Yandex began to streamline search results using a ranking formula for a specific region. At the same time, we introduced our own Matrixnet machine learning method , which is now used not only in search technologies in Yandex. For example, it is used by Seismotech to process seismic data and CERN to analyze meson decay . In 2011, Crypta appeared, which is also based on the MatrixNet. It is used in advertising technologies, automatically calculating gender, age and other characteristics of a person.

    But not all people's interests can be described in the questionnaire format, so we began to personalize the issuance, adjusting to the short-term and long-term interests of the user. For this, search platforms Kaliningrad and Dublin were launched . These are actually technologies for adapting a resource for a user, which we use primarily in the search, changing the issuance for a specific person who made a request.

    image

    Adapting the Internet to the interests of the user is a natural process. And the sites began to face problems similar to those that arose with us. What to show the user when a lot of content? What advice should I give when the content on each site is approximately the same? People still choose services that solve their problem faster, and remember those that gave a good recommendation.

    It’s good if the site has at least geo-targeting and is adapted to the user's region. Some begin to figure out the interests of the user, clearly cheating on them. For example, sports.ru asks each new user a question about their favorite football team.

    The most difficult step that most have yet to take is working with a large number of factors and fuzzy signals of interest. The market is already approaching the need for this step. The Atom program, which uses big data processing and machine learning technologies, can help make it.

    We can implement such things, thanks to a combination of two rather unique factors. On the one hand, we have a huge amount of knowledge about users and their behavior on the Internet. On the other hand, the ability to store and process big data in real time. We created these technologies for ourselves, because without them the search company cannot make a quality product. And their combination is probably in five companies in the world. We are the only ones who are ready to declare their desire to make such technologies a public product.

    Why do we announce our program without telling how we applied it on our services? If we did it for ourselves, it is obvious that we would announce it at a different stage of readiness. But we are talking about a thing whose rules of existence must be determined by all together.

    The discovery of the Atom is somewhat similar to the discovery of the Islands . While working on the integration of service responses in the Search, at some point we realized that to solve the problems of users together with site ownersmore promising. That is how the Islands appeared. When we worked on personalization, it came to the understanding that you cannot make a tool for solving problems personal, and leave the solutions, services on the Internet, not personalized. So the Atom appeared. Today we tell you about the Atom in order to start doing it together.

    image

    Both the Islands and the Atom are based on an understanding of changes on the Internet: it is more and more connected with reality, with the tasks of people. The transition from documents to services has already happened. The islands approach this on the one hand - they accompany a person on a task. Atom is on the other: it will allow the Internet to work with a specific person, and not with abstraction. Obviously, this helps to solve its problems, and in a broad sense: both transactional and those in which a person is primarily interested in content.

    In both products, we fundamentally change the approach of creation. As a metaphor, we always use crowdsourcing: the scale of the changes is too large, and for everything to work well, organically, the ecosystem needs the contribution of all its actors, all participants in the process.

    The use of our technologies by other sites will make it possible to make resources more appropriate for an individual person - so that they become more convenient and interesting for him. Talking about Atom on YaC, we proposed several key scenarios for which such an adaptation can go: changing the start page for a category of users, ranking objects within a category, smart recommendations based on a combination of ranking for a user and knowledge of the subject area the site is dedicated to. This is something like reference points - what they will develop into depends a lot on what webmasters working with sites will see in Atom.

    image

    The result that we see — moreover, it seems inevitable — is the ability to adapt to a specific person who has accessed a resource that has appeared on the entire Internet. This will bring more convenience to the user and give him interesting content. The site, accordingly, will be able to increase the conversion and viewing depth. If you are talking about transactions, a person will spend fewer clicks getting to the goal. If you have a content site - a person will not see boring materials on it, it is more likely to stick, with a significantly larger one - it will stick for a long time. If you are a smart store, thanks to Atom, you will get more and give more to the user: he will quickly find the phone that suits him, and in the recommended products he will see an unobvious, but cute cover for him.

    Today it is strange to see a large site that does not display correctly in different browsers. And ten years ago it was commonplace. In the end, it is strange to see a service that does not support login via social networks - this was not five years ago. Three or four years, and a person is simply confused, not having met the usual level of convenience on his next site.



    Thanks to Atom, everyone will be able to see content that, all other things being equal, would pass by him. The recommendatory component “pulls out” for a person more than what he himself was already looking for and found on the site: this will not narrow, but will expand his interests. The user will see only those products or the information that he needs, is interesting and useful, and for which he is ready to spend money or time.

    But it is hardly reasonable to assume that a person will be interested in all the new information. We will show him one that might be of interest to him. This is especially evident in examples with objects from the real world: a person was not going to find any content, but we realized that it would be right for him to show it. The atom does not close the man in what he already knows, but discovers from the unknown that he will like.

    It is impossible to show a person the entire content of the world, and it’s useless: the housewife doesn’t want to read about Warcraft, and the system administrator does not want to read about rhinestones. But within the framework of what a person does not see, there is something that would be more or less interesting to him. We help to lift upward what has a chance to hook it - and exit the current bubble.

    If you want to expand the sphere of interests to where it will be understood and accepted by man, Atom will help you. If you think that you can convince a person of the need for something that he really does not need, or lead him to a place where he really does not need it, you can rely on zombies, a whip or other strange tools. 

    All the technologies that we share will be based on Yandex's key expertise - working with data: machine learning, data extraction, creating factors , assessing the statistical reliability of the results, the pipeline of experiments.

    Currently, the technology is being tested on several Yandex services and with a number of external partners. The number of external tests will be increased, and when we get stable results and understand that the technology is ready for scaling, we will return to you with elaborated rules of participation. If you are ready to share your vision of using Atom or to join our experiments, write to us at atom-experiments@yandex-team.ru.

    Also popular now: