November 12, 2010 at 22:36What should we build Google
Today, as part of the Google Developers Day event , I managed to talk with Vladimir Ofitserov , who flew to Moscow specifically for the event. Vladimir is a specialist in the Google search quality department and I asked him a number of thematic questions that were interesting to the users of the site.
Let me remind you that since 1999 Vladimir Ofitserov worked at Inktomi
- One of the first Internet search engines, which became the basis of the search engine Yahoo. At Yahoo, Vladimir participated and led projects that were aimed at improving the ranking algorithm, scanning infrastructure and indexing. In 2008, Vladimir moved to the California-based Yandex Labs with a group of engineers from Yahoo, where he worked on projects aimed at improving search on the Russian Internet. Since 2010, she has been working in the Google Search Quality Division.
- Yahoo, Yandex, Google ... Vladimir, looking at such a sequence of places of your work, one involuntarily begs the question - how is this possible at all? Is it really so easy to take and move from the first company with all its secrets (NDA) to the second, which is the main competitor of the first? Tell us about the reasons for moving from one place to another.
- In 1999, one of my colleagues at Inktomi was Arkady Borkovsky , a friend of Arkady Volozh , with whom in 1997 they organized Yandex. For a long time I worked with Arkady at Inktomi, later at Yahoo, and when Yandex decided to open an office, the choice of the head of this office naturally fell to Arkady Borkovsky. Since I worked with them for a long time, I just became one of the people who founded this office. And about the secrets - in California from the point of law, there are no problems with this.
- That is, in fact, there was no enticement from one company, as such?
- Exactly. Yahoo, as you know, stopped doing searches and completely switched to Bing. This trend was noticeable some time ago - the company did not invest either in people, or in technology, or in computers ... And, in fact, leaving Yahoo simply lasted quite a long time. Yandex is a very good company (in terms of management, decision making, etc.), but when it is possible to move from a company that is searched by millions of forty people to a company that is searched by more than 500 million, the choice remains obvious. Especially if there is motivation to make the search really better.
- You have been working with Google since January of this year. Compared to previous jobs, are there any fundamental differences? Is there such a thing that a new place lacks something that was in previous places of work?
- The first thing you see when you come to Google is that it is a global company. People communicate much easier with each other, between offices, departments ... and communication is very dynamic. Yandex, nevertheless, grew up in Russia and this is felt - it is more closed ... he has less experience and approaches to organizing a global business.
- Now let's talk a little about the technical side of the issue. Why did Google decide to switch to a live search? And recently appeared site preview in search results.
- To begin with, about 30% of users do not see a live search at all - as a rule, those who type a request in the address bar of the browser. For the remaining average users, the "live search" saves about two to three seconds. But we must not forget about people who type text with two fingers slowly - for them, a “live search" saves from 30 seconds or more. By clicking only a few letters, they are very likely to be able to select the query they need, which will no longer have to be printed. We decided that for many this could be a very big plus, and everyone else can always turn off this feature.
- It seems to me that in most cases the user in general terms imagines what he is looking for - accordingly, he simply does not need all the side effects. But is there, for example, such that in these "invisible results" spend money advertisers using Google AdWords?
- As for advertising, there is no problem with this - money begins to be spent only when a person has completed his request and there are mechanisms for determining this moment. Conversely, in the case of a live ad search, less is shown.
- Could it happen that with such a prediction of requests there will be an indirect manipulation of the person conducting the search? When he was just thinking of looking for something, and he was already helpfully offered what he “wants” to look for or what someone already paid for as “what to look for” (“everyone is looking for it!” Or “everyone consider this an answer! ”).
- The key factor that the search engine offers you to complete some kind of query is that we know that the result on it will be “good”. And if the user among the proposed options sees similar to the one he wants to find, then from the point of view of the search engine this is the ideal option to formulate a query, the goal. That is, if the user has difficulty with the wording of the request, but in the list offered by the machine he sees something similar to what he wants to find, then he will probably find it much faster.
- What can you say about social search? When does Google plan to create a search that determines the relevance of search results based on interaction and user assistance, contrary to common search methods based on algorithms?
- I believe that such a search will appear in the very near future, next or even this year. Moreover - now it is possible to specify your twitter-, facebook-account (and some other social networks) in the Google profile, after which Google will build a chain of your friends. And if one of them publishes some link on his blog or on a social network, and the information content of this link will be similar to your search query, then you will get a kind of “friendly” result with the corresponding mark.
- One of the problems of the current search (of any) is that he is looking not “in depth”, but “in breadth”. That is, for example, at the request of "pen" in the results will be a huge number of a wide variety of handles - from writing to door. But at the same time, “at the request of a pen, 1,900,000 results were found” - it’s the same as asking someone “What is a pen?” and get the answer: “I know more than 1,900,000 answers to this question, what’s yours?” That is, all instant gains are lost against the background of heterogeneity of information. Is there any way to deal with this?
- At the moment, the machine really can not predict at such a level what exactly you are looking for and mean, especially when it comes to such homonyms. However, for such situations, we try to display information about several values. Well, technology, as we see, does not stand still - I think nothing is impossible.
- What other problems exist in the modern search? What does not suit you personally at the moment and what would you like to correct as soon as possible?
- There are a lot of problems, but one of the most serious, I think, is spam in the search results. At least for the Russian market this is one of the key factors and we will try to filter this information.
Another trend that has been seen for several years is the provision of information without the need to press a large number of keys. For example, one of the steps to solving this problem was the recently introduced voice search.
- How do you see the search, for example, after 5 years? In 10? Now they talk a lot about augmented reality ...?
“ If only I knew ... ” said Vladimir and laughed. At this time, in a whisper, a hint was heard from the outside, from a colleague: “ I would then invest! "
Perhaps the search will become even closer to the user. Personalization, localization, regionalization, socialization - I think all of these aspects will significantly change the current search engine. For the reason that there is a lot of information in this data - no one will say more for you than your friends.
- Accordingly, the recent conflict with Facebook is actually much more significant than it might seem at first glance?
- We always take competitors very seriously and, I think, do the same. But between search and social networks there are a lot of common ground, thanks to which the work of these services can be made even better. The quality of our issuance could be much higher if we knew more about the “likes” that Facebook placed throughout the Internet.
- That is, now, in fact, it is this competition with Facebook that is currently slowing down the search progress?
- Exactly. I hope that sooner or later these barriers will become much lower.
- Vladimir, for more than 10 years of “search” work, you probably have accumulated many interesting or unusual stories? Tell something from the most memorable, so to speak, a couple of tales to laugh and cry a couple of rakes.
- A year ago, in 2002, there was a funny case when I worked on a runtime-system (a system that should respond to requests within half a second), I found a search query in its logs, which was processed for more than a minute, I decided to figure it out. It turned out that the user entered in the search line “i am alone in valentine's day” with a space after each letter (“I amalone I nvalent I ne 'sday”). As a result, the search engine was presented with an almost whole alphabet, where for each letter there were hundreds and thousands of documents - the system struggled to combine them and produce the most suitable result. [ note: I just tried to enter a similar query in Yandex - really a butthert, albeit not for a minute; in google ok ]
I've made a list of some fun search questions here ... sometimes it's fun to read.
- About titanium scrap in the train toilet? :)
- No, this is already a classic :) Here is something fresh. It is unclear what users are guided when looking for this. Well, for example, “ what fringe suits me? ".
As for the rake ... it was also full of everything. But from the point of view of the programmer, the most terrible rake is to free that memory that you do not own. It will not lead to anything good.
- Yeah, interesting. And what can you advise to those who, for example, want to follow in your footsteps? Maybe you can recommend some of the most interesting books ... that is, not just Google for Dummies, but something more serious, really worthwhile. Or is there nothing interesting in printed form and is it worth looking for the most delicious in the vast expanses of the network?
- Walking in my footsteps is not at all necessary - everyone should choose their own path, really interesting. Well, as far as information is concerned, I can recommend two interesting books: Introduction to Information Retrieval [Christopher D. Manning, Prabhakar] is a more academic book written by a professor from Stanford. So to speak, for the basics. A " Search Engines: Information Retrieval in Practice"[Bruce Croft, Donald Metzler, and Trevor Stohman] is already a hands-on book written by Google engineers. It is suitable for those who, for example, want to make their own search engine - it tells about the work of many mechanisms, about writing effective code and many other useful things. And, of course, the Internet - if you wish, you can always find interesting, and most importantly, relevant information there.
Thanks! And finally, maybe share some secret? Anything straightforward, exclusive to readers of our site?
- Well, they are secrets and secrets :) Nevertheless, something is periodically revealed - like, for example, a programming language recently opened to the public, designed to process VERY LARGE volumes (more than the index!) Of logs. It was written by one Russian (I’m not afraid to say this word to scientists) and for a long time was our “know-how”, but about a week ago this information became public; you can google if you wish. Well, or the same Closure and GWT for developing sufficiently rich Java web applications (which are then compiled in JavaScript and packaged so that they are optimal for compilation, loading, and runtime in browsers) - all this Google has made available to developers.
We have Brad Fitzpatrick, the creator of LiveJournal - in fact, he is an “ordinary” programmer and at the time, in addition to LJ, wrote a few things for Google (mostly methods and classes) that are still relevant today - they literally work on them now all and similar examples can be given quite a lot. I can say more simply - Google is the company that makes religion out of engineering by creating projects “forever”. Most often, these are much more complex mechanisms (than those of competitors), all the subtleties of which cannot be taken and opened.
At this our meeting came to an end, I took some photos as a keepsake and went to the event itself - I was pleasantly surprised by the large number of people registered on Habré.
Good luck!
- One of the first Internet search engines, which became the basis of the search engine Yahoo. At Yahoo, Vladimir participated and led projects that were aimed at improving the ranking algorithm, scanning infrastructure and indexing. In 2008, Vladimir moved to the California-based Yandex Labs with a group of engineers from Yahoo, where he worked on projects aimed at improving search on the Russian Internet. Since 2010, she has been working in the Google Search Quality Division.
- Yahoo, Yandex, Google ... Vladimir, looking at such a sequence of places of your work, one involuntarily begs the question - how is this possible at all? Is it really so easy to take and move from the first company with all its secrets (NDA) to the second, which is the main competitor of the first? Tell us about the reasons for moving from one place to another.
- In 1999, one of my colleagues at Inktomi was Arkady Borkovsky , a friend of Arkady Volozh , with whom in 1997 they organized Yandex. For a long time I worked with Arkady at Inktomi, later at Yahoo, and when Yandex decided to open an office, the choice of the head of this office naturally fell to Arkady Borkovsky. Since I worked with them for a long time, I just became one of the people who founded this office. And about the secrets - in California from the point of law, there are no problems with this.
- That is, in fact, there was no enticement from one company, as such?
- Exactly. Yahoo, as you know, stopped doing searches and completely switched to Bing. This trend was noticeable some time ago - the company did not invest either in people, or in technology, or in computers ... And, in fact, leaving Yahoo simply lasted quite a long time. Yandex is a very good company (in terms of management, decision making, etc.), but when it is possible to move from a company that is searched by millions of forty people to a company that is searched by more than 500 million, the choice remains obvious. Especially if there is motivation to make the search really better.
- You have been working with Google since January of this year. Compared to previous jobs, are there any fundamental differences? Is there such a thing that a new place lacks something that was in previous places of work?
- The first thing you see when you come to Google is that it is a global company. People communicate much easier with each other, between offices, departments ... and communication is very dynamic. Yandex, nevertheless, grew up in Russia and this is felt - it is more closed ... he has less experience and approaches to organizing a global business.
- Now let's talk a little about the technical side of the issue. Why did Google decide to switch to a live search? And recently appeared site preview in search results.
- To begin with, about 30% of users do not see a live search at all - as a rule, those who type a request in the address bar of the browser. For the remaining average users, the "live search" saves about two to three seconds. But we must not forget about people who type text with two fingers slowly - for them, a “live search" saves from 30 seconds or more. By clicking only a few letters, they are very likely to be able to select the query they need, which will no longer have to be printed. We decided that for many this could be a very big plus, and everyone else can always turn off this feature.
- It seems to me that in most cases the user in general terms imagines what he is looking for - accordingly, he simply does not need all the side effects. But is there, for example, such that in these "invisible results" spend money advertisers using Google AdWords?
- As for advertising, there is no problem with this - money begins to be spent only when a person has completed his request and there are mechanisms for determining this moment. Conversely, in the case of a live ad search, less is shown.
- Could it happen that with such a prediction of requests there will be an indirect manipulation of the person conducting the search? When he was just thinking of looking for something, and he was already helpfully offered what he “wants” to look for or what someone already paid for as “what to look for” (“everyone is looking for it!” Or “everyone consider this an answer! ”).
- The key factor that the search engine offers you to complete some kind of query is that we know that the result on it will be “good”. And if the user among the proposed options sees similar to the one he wants to find, then from the point of view of the search engine this is the ideal option to formulate a query, the goal. That is, if the user has difficulty with the wording of the request, but in the list offered by the machine he sees something similar to what he wants to find, then he will probably find it much faster.
- What can you say about social search? When does Google plan to create a search that determines the relevance of search results based on interaction and user assistance, contrary to common search methods based on algorithms?
- I believe that such a search will appear in the very near future, next or even this year. Moreover - now it is possible to specify your twitter-, facebook-account (and some other social networks) in the Google profile, after which Google will build a chain of your friends. And if one of them publishes some link on his blog or on a social network, and the information content of this link will be similar to your search query, then you will get a kind of “friendly” result with the corresponding mark.
- One of the problems of the current search (of any) is that he is looking not “in depth”, but “in breadth”. That is, for example, at the request of "pen" in the results will be a huge number of a wide variety of handles - from writing to door. But at the same time, “at the request of a pen, 1,900,000 results were found” - it’s the same as asking someone “What is a pen?” and get the answer: “I know more than 1,900,000 answers to this question, what’s yours?” That is, all instant gains are lost against the background of heterogeneity of information. Is there any way to deal with this?
- At the moment, the machine really can not predict at such a level what exactly you are looking for and mean, especially when it comes to such homonyms. However, for such situations, we try to display information about several values. Well, technology, as we see, does not stand still - I think nothing is impossible.
- What other problems exist in the modern search? What does not suit you personally at the moment and what would you like to correct as soon as possible?
- There are a lot of problems, but one of the most serious, I think, is spam in the search results. At least for the Russian market this is one of the key factors and we will try to filter this information.
Another trend that has been seen for several years is the provision of information without the need to press a large number of keys. For example, one of the steps to solving this problem was the recently introduced voice search.
- How do you see the search, for example, after 5 years? In 10? Now they talk a lot about augmented reality ...?
“ If only I knew ... ” said Vladimir and laughed. At this time, in a whisper, a hint was heard from the outside, from a colleague: “ I would then invest! "
Perhaps the search will become even closer to the user. Personalization, localization, regionalization, socialization - I think all of these aspects will significantly change the current search engine. For the reason that there is a lot of information in this data - no one will say more for you than your friends.
- Accordingly, the recent conflict with Facebook is actually much more significant than it might seem at first glance?
- We always take competitors very seriously and, I think, do the same. But between search and social networks there are a lot of common ground, thanks to which the work of these services can be made even better. The quality of our issuance could be much higher if we knew more about the “likes” that Facebook placed throughout the Internet.
- That is, now, in fact, it is this competition with Facebook that is currently slowing down the search progress?
- Exactly. I hope that sooner or later these barriers will become much lower.
- Vladimir, for more than 10 years of “search” work, you probably have accumulated many interesting or unusual stories? Tell something from the most memorable, so to speak, a couple of tales to laugh and cry a couple of rakes.
- A year ago, in 2002, there was a funny case when I worked on a runtime-system (a system that should respond to requests within half a second), I found a search query in its logs, which was processed for more than a minute, I decided to figure it out. It turned out that the user entered in the search line “i am alone in valentine's day” with a space after each letter (“I amalone I nvalent I ne 'sday”). As a result, the search engine was presented with an almost whole alphabet, where for each letter there were hundreds and thousands of documents - the system struggled to combine them and produce the most suitable result. [ note: I just tried to enter a similar query in Yandex - really a butthert, albeit not for a minute; in google ok ]
I've made a list of some fun search questions here ... sometimes it's fun to read.
- About titanium scrap in the train toilet? :)
- No, this is already a classic :) Here is something fresh. It is unclear what users are guided when looking for this. Well, for example, “ what fringe suits me? ".
As for the rake ... it was also full of everything. But from the point of view of the programmer, the most terrible rake is to free that memory that you do not own. It will not lead to anything good.
- Yeah, interesting. And what can you advise to those who, for example, want to follow in your footsteps? Maybe you can recommend some of the most interesting books ... that is, not just Google for Dummies, but something more serious, really worthwhile. Or is there nothing interesting in printed form and is it worth looking for the most delicious in the vast expanses of the network?
- Walking in my footsteps is not at all necessary - everyone should choose their own path, really interesting. Well, as far as information is concerned, I can recommend two interesting books: Introduction to Information Retrieval [Christopher D. Manning, Prabhakar] is a more academic book written by a professor from Stanford. So to speak, for the basics. A " Search Engines: Information Retrieval in Practice"[Bruce Croft, Donald Metzler, and Trevor Stohman] is already a hands-on book written by Google engineers. It is suitable for those who, for example, want to make their own search engine - it tells about the work of many mechanisms, about writing effective code and many other useful things. And, of course, the Internet - if you wish, you can always find interesting, and most importantly, relevant information there.
Thanks! And finally, maybe share some secret? Anything straightforward, exclusive to readers of our site?
- Well, they are secrets and secrets :) Nevertheless, something is periodically revealed - like, for example, a programming language recently opened to the public, designed to process VERY LARGE volumes (more than the index!) Of logs. It was written by one Russian (I’m not afraid to say this word to scientists) and for a long time was our “know-how”, but about a week ago this information became public; you can google if you wish. Well, or the same Closure and GWT for developing sufficiently rich Java web applications (which are then compiled in JavaScript and packaged so that they are optimal for compilation, loading, and runtime in browsers) - all this Google has made available to developers.
We have Brad Fitzpatrick, the creator of LiveJournal - in fact, he is an “ordinary” programmer and at the time, in addition to LJ, wrote a few things for Google (mostly methods and classes) that are still relevant today - they literally work on them now all and similar examples can be given quite a lot. I can say more simply - Google is the company that makes religion out of engineering by creating projects “forever”. Most often, these are much more complex mechanisms (than those of competitors), all the subtleties of which cannot be taken and opened.
At this our meeting came to an end, I took some photos as a keepsake and went to the event itself - I was pleasantly surprised by the large number of people registered on Habré.
Good luck!