THERE WILL BE NO MONEY IN THE INDUSTRY UNTIL THE VALUE OF TELEVISION RETURNS
VITALY VASILIEV, EPG SERVICE: "THERE WILL BE NO MONEY IN THE INDUSTRY UNTIL THE VALUE OF TELEVISION RETURNS"
Mikhail Grigoryev, Telesputnik
The EPG system (Electronic Program Guide, or electronic TV guide, - ed.) as a framework for building a television service can reduce the time to search for content. Together with the connected recommendation system, it will make TV viewing active and help the user realize what he pays the operator for. This was told by the general director of the EPG Service company Vitaly Vasiliev.
Mikhail Grigoryev: Many users imagine EPG simply as a program guide that is built into any pay-TV service. But that's not quite true. Explain please what else a modern EPG service can do.
Vitaly Vasiliev: First you need to decide what an #EPG is. Everyone has a different idea about this. I would single out three EPG classes. The first is when the user remembers the buttons on the remote control, which, when pressed, call the #TV channel he needs. In this case, it doesn't even use EPG functions. The second is the classic TV guide, or grid, which is presented by #DVB-broadcasting operators. It has limited functions, a minimum of text, allows you to scroll through channels and determine what is going on and what will go on the air. The third class is more modern: as a rule, it is used by #IPTV and #OTT operators. It contains all the functions, including illustrations for programs, navigation system, content collections. With the help of such an EPG, content recording mechanisms are implemented for the deferred viewing service, which is actively used by viewers today. The further we develop, the more functions the EPG service has.
Mikhail Grigoryev:According to a study by Ericsson ConsumerLab, Russians spend 36 minutes a day searching for TV #content. Are EPG services able to reduce this time?
Vitaly Vasiliev: It would be interesting to compare this figure with the time spent by a user searching for content on YouTube. In this case, we would be able to answer the question posed. And also - compare the search functions that this video platform has with those that TV #platforms have. Personally, I don't need to search for anything on YouTube: I open the feed — and it already has the content I'm interested in.
MG: But after all, the proposed content does not always suit the interests of the user.
Vitaly Vasiliev: We need to make objective studies. Everyone has a different experience using YouTube. The platform correctly recommends content to some, but not to others. Let's go back to modern pay-TV platforms. As a rule, the fewer functions they have, the more time the viewer spends searching. Conversely, the more features and features for navigation, the less time it takes to search. This is a direct relation.
Mikhail Grigoryev: I can't call YouTube or Netflix services with a low set of functions. Nevertheless, we often hear from experts that their recommendation systems are far from ideal.
Vitaly Vasiliev: Why are we talking about these platforms? Because most users are sitting there. And the growth of the subscriber base occurs primarily with them. The #pay_TV operator would not refuse such a number of users that the same YouTube has. When we talk about the platform, we mean not only a system of recommendations, but also all the mechanisms that are involved in keeping the user's attention. The recommendation system may work correctly, it may work incorrectly, someone does not use it at all. But the fact remains: everyone uses YouTube. And one of the reasons is that it has features that make it easy to find the right content. Let's recall the same search bar where there is a relevant search. Do you know many platforms that have the same capability? Or platforms that have a mobile application from which you can connect to the TV and control playback?
The recommendation system is important and necessary, but it is not the only thing that allows you to keep the viewer's attention. It's also marked-up content that you can search for, and collections of content, and subscriptions, and content filters. If we want pay TV to develop, all this wealth of features must be in it.
Returning to the previous question about reducing content search time - yes, EPG #services as a framework for building a television service are able to do this. It is impossible to build a service in demand by users without data and metadata for EPG, without graphic design of content.
Mikhail Grigoryev: Tell us, please, how your recommendation system works.
Vitaly Vasiliev: It has several components. The first is a collaborative recommendation system that groups users and determines by group which content may be of interest to them. It is complemented by the second component - the associative approach, when content similar to what the user likes is determined. The third component is the adjustment of editorial collections according to certain principles. It allows you to make a selection, for example, of Russian comedies or documentaries about tigers. This is purely editorial work, where you need to select hype topics based on user experience, find the appropriate user segments — and make such collections of content for them.
Mikhail Grigoryev: What should the specialists who make editorial selections be able to do?
Vitaly Vasiliev:These are product analysts who study the viewing of content by users, identify groups of people with similar needs and know what they can recommend. This is the so-called data-driven approach (a data-driven approach to management - ed.).
Mikhail Grigoryev: Is it possible in your system to set various parameters, such as user profiles, recommendations depending on the time of day, and so on?
Vitaly Vasiliev: We have an analytics mechanism that can determine the viewer by the nature of viewing content. One TV-set can be used by different people, each of whom consumes their own content. And the recommendation system understands this as different users, offering each of them the most relevant content. Although the user profile is unknown, that is, the recommendation system does not know that, for example, mom watches TV during the day and dad watches it in the evening.
Mikhail Grigoryev: What is the fundamental difference for the end user of the pay-TV platform, is the #recommendation system included or not in the service?
Vitaly Vasiliev: If we consider the recommendation system as a whole, then the user cannot see all the content. The user has no experience and does not know what programs are on television. He's used to turning on certain channels. And if there is no content he needs on the air, then he spends his leisure time not with the help of TV. The recommendation system allows you to make TV viewing active, get involved in it, understand what the user pays the operator for.
Mikhail Grigoryev: It has been repeatedly stated that expert recommendations of content are much more accurate than those offered by artificial intelligence. Can we find ourselves in a situation where AI will catch up with live experts in terms of efficiency?
Vitaly Vasiliev: I think so, but it will be a completely different television. There are probably ten years left to wait.
Mikhail Grigoryev: Why in ten years, and not earlier?
Vitaly Vasiliev: Artificial intelligence is already able to draw pictures. But with the use of AI in the television sphere, unfortunately, not everything is so smooth. Until the value of #television returns, there will be no money in this industry. And if there is no money— there will be no technology. In order for this to appear, you need to attract the user, make a valuable product for him. Nine years ago, EPG Service came up with the idea of using metadata. Five years ago, we offered operators products that they have only just begun to implement. In other words, only now the market has come to understand that this needs to be done, otherwise no one will need television.
Mikhail Grigoryev: What prevented you from understanding this earlier?
Vitaly Vasiliev: Five years ago, at CSTB, I ran from stand to stand and offered operators the technology of using broad metadata that allows you to build a system of recommendations and make collections of content for users using filters. The operators reacted with interest, but replied that they needed a ready-made solution. The developers, in turn, said that they were also interested in this, but there was no demand among operators. It turned out to be a vicious circle. The lack of a community uniting solution developers and pay-TV #operators is one of the industry's problems. Another problem is that pay TV is a very inert sphere. Everything is ready for the active introduction of artificial intelligence on television, only money is missing. But if the operators consolidate, then everything will work out.
Mikhail Grigoryev: Which #recommendation system do you think is optimal — one based only on editorial selections, or one that combines them with artificial intelligence?
Vitaly Vasiliev: Netflix answered this question. First, they developed a recommendation system based on artificial intelligence, and now they have hired editors who mark up content with metadata and make editorial selections. Because Netflix realized that no matter how smart an AI is, it will not surpass a human. At the same time, they do not discount the system of recommendations based on artificial intelligence itself and continue to actively develop it. One technique complements the other. I think that the result will be a symbiosis of machine and human efforts.
Mikhail Grigoryev: One of your services is the mapping of linear TV and video on demand libraries (Video on Demand, #VoD, - ed.). What does this give and how is it related to the work of the recommendation system?
Vitaly Vasiliev: As a rule, pay-TV operators have a video-on-demand library and linear TV with EPG. The content in them may be similar and may even overlap. The mapping function is data that makes it clear that the content on linear TV is the same as the content in the #VoD library. Let's say a user watched the first and second series of a TV show and wanted to see the third. It will be broadcast on television in a week, and the operator is ready to offer to buy the entire series in the VoD library. The opposite situation also happens: the series appears on television earlier than in the VoD library.
Mapping is a recommendation system that works both ways. Based on the experience of viewing a particular user in the VoD library, it allows you to understand what you can recommend from linear content, and vice versa. If the metadata in the #VoD library and linear TV are different, then we will not be able to do this, because for us they will be different, unrelated systems. Mapping helps to connect the content of video on demand and linear TV and, based on this, build a unified system of recommendations, install filters and selections, and as a result, increase the marginality and efficiency of the operator's business.
Mikhail Grigoryev: What else helps you develop a recommendation system?
Vitaly Vasiliev: The trend is that the user does not want to run to the TV so as not to miss a program or a series. On the contrary, he wants television to adapt to his time and interests. To do this, catch up technology has appeared (with the help of the Internet, it allows you to watch TV #programs after they go on the air, - ed.). But there is a problem with viewing the content in the recording — inaccurate positioning of the transmission start. It takes place on the basis of the EPG system provided by TV channels, and there the accuracy is plus or minus five minutes. In addition, there are shifts and replacements of programs that the channels do not report in advance. As a result, when a user wants to see the content he needs, he stumbles upon the wrong thing. To solve this problem, we use precise positioning in the EPG system: for each broadcast event in the archive, time is given with an accuracy of plus or minus three seconds. This is enough for comfortable viewing of TV programs.
There is one more point. If you build a recommendation system based on raw data from a TV channel, without precise markup, then it will recommend content to the viewer based on viewing a program that he did not intend to watch. Correct positioning helps to reduce such errors and increase the accuracy of the recommendation system.
Mikhail Grigoryev: EPG Service is an IT company: programmers and developers work for you. In which areas do you experience a shortage of specialists the most?
First of all, we lack IT specialists. Since the beginning of the pandemic, the market has changed a lot and all IT specialists have been dismantled. On the other hand, the prices for the services of IT specialists have doubled. If earlier remoting was a business strategy for us and we offered programmers the opportunity to work outside the office, now it has become the norm for everyone, including large companies. For example, if before a programmer writing in Python could be found, conditionally, for 150 thousand rubles, today - for at least 250 thousand rubles.
Mikhail Grigoryev: How would you like to see your recommendation system in the future? What should she be able to do?
Vitaly Vasiliev: It should become more effective from the point of view of the metrics accepted in the market. Of course, one hundred percent efficiency cannot be achieved either with the help of artificial intelligence or with the help of editorial collections, but if it turns out to do better than on YouTube, then we can assume that the task has been solved. In this case, television will find a new life.
Mikhail Grigoryev, Telesputnik
The EPG system (Electronic Program Guide, or electronic TV guide, - ed.) as a framework for building a television service can reduce the time to search for content. Together with the connected recommendation system, it will make TV viewing active and help the user realize what he pays the operator for. This was told by the general director of the EPG Service company Vitaly Vasiliev.
Mikhail Grigoryev: Many users imagine EPG simply as a program guide that is built into any pay-TV service. But that's not quite true. Explain please what else a modern EPG service can do.
Vitaly Vasiliev: First you need to decide what an #EPG is. Everyone has a different idea about this. I would single out three EPG classes. The first is when the user remembers the buttons on the remote control, which, when pressed, call the #TV channel he needs. In this case, it doesn't even use EPG functions. The second is the classic TV guide, or grid, which is presented by #DVB-broadcasting operators. It has limited functions, a minimum of text, allows you to scroll through channels and determine what is going on and what will go on the air. The third class is more modern: as a rule, it is used by #IPTV and #OTT operators. It contains all the functions, including illustrations for programs, navigation system, content collections. With the help of such an EPG, content recording mechanisms are implemented for the deferred viewing service, which is actively used by viewers today. The further we develop, the more functions the EPG service has.
Mikhail Grigoryev:According to a study by Ericsson ConsumerLab, Russians spend 36 minutes a day searching for TV #content. Are EPG services able to reduce this time?
Vitaly Vasiliev: It would be interesting to compare this figure with the time spent by a user searching for content on YouTube. In this case, we would be able to answer the question posed. And also - compare the search functions that this video platform has with those that TV #platforms have. Personally, I don't need to search for anything on YouTube: I open the feed — and it already has the content I'm interested in.
MG: But after all, the proposed content does not always suit the interests of the user.
Vitaly Vasiliev: We need to make objective studies. Everyone has a different experience using YouTube. The platform correctly recommends content to some, but not to others. Let's go back to modern pay-TV platforms. As a rule, the fewer functions they have, the more time the viewer spends searching. Conversely, the more features and features for navigation, the less time it takes to search. This is a direct relation.
Mikhail Grigoryev: I can't call YouTube or Netflix services with a low set of functions. Nevertheless, we often hear from experts that their recommendation systems are far from ideal.
Vitaly Vasiliev: Why are we talking about these platforms? Because most users are sitting there. And the growth of the subscriber base occurs primarily with them. The #pay_TV operator would not refuse such a number of users that the same YouTube has. When we talk about the platform, we mean not only a system of recommendations, but also all the mechanisms that are involved in keeping the user's attention. The recommendation system may work correctly, it may work incorrectly, someone does not use it at all. But the fact remains: everyone uses YouTube. And one of the reasons is that it has features that make it easy to find the right content. Let's recall the same search bar where there is a relevant search. Do you know many platforms that have the same capability? Or platforms that have a mobile application from which you can connect to the TV and control playback?
The recommendation system is important and necessary, but it is not the only thing that allows you to keep the viewer's attention. It's also marked-up content that you can search for, and collections of content, and subscriptions, and content filters. If we want pay TV to develop, all this wealth of features must be in it.
Returning to the previous question about reducing content search time - yes, EPG #services as a framework for building a television service are able to do this. It is impossible to build a service in demand by users without data and metadata for EPG, without graphic design of content.
Mikhail Grigoryev: Tell us, please, how your recommendation system works.
Vitaly Vasiliev: It has several components. The first is a collaborative recommendation system that groups users and determines by group which content may be of interest to them. It is complemented by the second component - the associative approach, when content similar to what the user likes is determined. The third component is the adjustment of editorial collections according to certain principles. It allows you to make a selection, for example, of Russian comedies or documentaries about tigers. This is purely editorial work, where you need to select hype topics based on user experience, find the appropriate user segments — and make such collections of content for them.
Mikhail Grigoryev: What should the specialists who make editorial selections be able to do?
Vitaly Vasiliev:These are product analysts who study the viewing of content by users, identify groups of people with similar needs and know what they can recommend. This is the so-called data-driven approach (a data-driven approach to management - ed.).
Mikhail Grigoryev: Is it possible in your system to set various parameters, such as user profiles, recommendations depending on the time of day, and so on?
Vitaly Vasiliev: We have an analytics mechanism that can determine the viewer by the nature of viewing content. One TV-set can be used by different people, each of whom consumes their own content. And the recommendation system understands this as different users, offering each of them the most relevant content. Although the user profile is unknown, that is, the recommendation system does not know that, for example, mom watches TV during the day and dad watches it in the evening.
Mikhail Grigoryev: What is the fundamental difference for the end user of the pay-TV platform, is the #recommendation system included or not in the service?
Vitaly Vasiliev: If we consider the recommendation system as a whole, then the user cannot see all the content. The user has no experience and does not know what programs are on television. He's used to turning on certain channels. And if there is no content he needs on the air, then he spends his leisure time not with the help of TV. The recommendation system allows you to make TV viewing active, get involved in it, understand what the user pays the operator for.
Mikhail Grigoryev: It has been repeatedly stated that expert recommendations of content are much more accurate than those offered by artificial intelligence. Can we find ourselves in a situation where AI will catch up with live experts in terms of efficiency?
Vitaly Vasiliev: I think so, but it will be a completely different television. There are probably ten years left to wait.
Mikhail Grigoryev: Why in ten years, and not earlier?
Vitaly Vasiliev: Artificial intelligence is already able to draw pictures. But with the use of AI in the television sphere, unfortunately, not everything is so smooth. Until the value of #television returns, there will be no money in this industry. And if there is no money— there will be no technology. In order for this to appear, you need to attract the user, make a valuable product for him. Nine years ago, EPG Service came up with the idea of using metadata. Five years ago, we offered operators products that they have only just begun to implement. In other words, only now the market has come to understand that this needs to be done, otherwise no one will need television.
Mikhail Grigoryev: What prevented you from understanding this earlier?
Vitaly Vasiliev: Five years ago, at CSTB, I ran from stand to stand and offered operators the technology of using broad metadata that allows you to build a system of recommendations and make collections of content for users using filters. The operators reacted with interest, but replied that they needed a ready-made solution. The developers, in turn, said that they were also interested in this, but there was no demand among operators. It turned out to be a vicious circle. The lack of a community uniting solution developers and pay-TV #operators is one of the industry's problems. Another problem is that pay TV is a very inert sphere. Everything is ready for the active introduction of artificial intelligence on television, only money is missing. But if the operators consolidate, then everything will work out.
Mikhail Grigoryev: Which #recommendation system do you think is optimal — one based only on editorial selections, or one that combines them with artificial intelligence?
Vitaly Vasiliev: Netflix answered this question. First, they developed a recommendation system based on artificial intelligence, and now they have hired editors who mark up content with metadata and make editorial selections. Because Netflix realized that no matter how smart an AI is, it will not surpass a human. At the same time, they do not discount the system of recommendations based on artificial intelligence itself and continue to actively develop it. One technique complements the other. I think that the result will be a symbiosis of machine and human efforts.
Mikhail Grigoryev: One of your services is the mapping of linear TV and video on demand libraries (Video on Demand, #VoD, - ed.). What does this give and how is it related to the work of the recommendation system?
Vitaly Vasiliev: As a rule, pay-TV operators have a video-on-demand library and linear TV with EPG. The content in them may be similar and may even overlap. The mapping function is data that makes it clear that the content on linear TV is the same as the content in the #VoD library. Let's say a user watched the first and second series of a TV show and wanted to see the third. It will be broadcast on television in a week, and the operator is ready to offer to buy the entire series in the VoD library. The opposite situation also happens: the series appears on television earlier than in the VoD library.
Mapping is a recommendation system that works both ways. Based on the experience of viewing a particular user in the VoD library, it allows you to understand what you can recommend from linear content, and vice versa. If the metadata in the #VoD library and linear TV are different, then we will not be able to do this, because for us they will be different, unrelated systems. Mapping helps to connect the content of video on demand and linear TV and, based on this, build a unified system of recommendations, install filters and selections, and as a result, increase the marginality and efficiency of the operator's business.
Mikhail Grigoryev: What else helps you develop a recommendation system?
Vitaly Vasiliev: The trend is that the user does not want to run to the TV so as not to miss a program or a series. On the contrary, he wants television to adapt to his time and interests. To do this, catch up technology has appeared (with the help of the Internet, it allows you to watch TV #programs after they go on the air, - ed.). But there is a problem with viewing the content in the recording — inaccurate positioning of the transmission start. It takes place on the basis of the EPG system provided by TV channels, and there the accuracy is plus or minus five minutes. In addition, there are shifts and replacements of programs that the channels do not report in advance. As a result, when a user wants to see the content he needs, he stumbles upon the wrong thing. To solve this problem, we use precise positioning in the EPG system: for each broadcast event in the archive, time is given with an accuracy of plus or minus three seconds. This is enough for comfortable viewing of TV programs.
There is one more point. If you build a recommendation system based on raw data from a TV channel, without precise markup, then it will recommend content to the viewer based on viewing a program that he did not intend to watch. Correct positioning helps to reduce such errors and increase the accuracy of the recommendation system.
Mikhail Grigoryev: EPG Service is an IT company: programmers and developers work for you. In which areas do you experience a shortage of specialists the most?
First of all, we lack IT specialists. Since the beginning of the pandemic, the market has changed a lot and all IT specialists have been dismantled. On the other hand, the prices for the services of IT specialists have doubled. If earlier remoting was a business strategy for us and we offered programmers the opportunity to work outside the office, now it has become the norm for everyone, including large companies. For example, if before a programmer writing in Python could be found, conditionally, for 150 thousand rubles, today - for at least 250 thousand rubles.
Mikhail Grigoryev: How would you like to see your recommendation system in the future? What should she be able to do?
Vitaly Vasiliev: It should become more effective from the point of view of the metrics accepted in the market. Of course, one hundred percent efficiency cannot be achieved either with the help of artificial intelligence or with the help of editorial collections, but if it turns out to do better than on YouTube, then we can assume that the task has been solved. In this case, television will find a new life.