Model of domestic video sites: Netflix and its precise recommendation algorithm

nowadays, domestic Internet streaming service has become a mohican. Money for copyright, confess dad, such as acquisition, is undoubtedly the domestic video industry development present situation. In the face of the new wave of technology, traditional video service providers, the sense of insecurity focus completely on the hype, few people can think calmly and user habits change how to deal with industry. Facts have proven that focus on content production and distribution, provide accurate recommendation mechanism, multiple platforms seamless docking, perhaps is the tao of the survival and development of video website.

, by contrast, foreign colleagues seem to do better than us. The following is a wired streaming service Netflix’s web site to the United States. The company’s two senior executives emphasized how Netflix by upgrading algorithm, to realize the precise recommendation mechanism. Hunting cloud network full text compiling, the hope can for the tumulfuous furnituer industry, bring some valuable rational thinking.

if you are in the last century 60 s interested in star trek, so it can automatically for you recommend “mission: impossible” series of the same type; If you like is the doctor who, so it can also be served for you English version of the ones the heart; If above all can not lift spirit make you feel, and you just love to see the fight a dusk of this type, so it will be close to in your personal home page, add a line category labels: “violence has a visual impact, adventure films.

first to sell a imprison son, look at the following these tips, can you guess its name?

this is a famous video website in the United States, headquartered in silicon valley, currently has 800 professional engineers;

in addition to selling the original DVD business, it is mainly to provide users with online subscription video watching services;

with other types of sites, it has a strong recommendation mechanism, by gathering user using relevant data, can achieve precise push;

it is understood that the company sold a total of 4 billion DVDS, and the first quarter of this year it provides up to 4 billion hours of streaming media service.

I think you have guessed, it is the Netflix.

in recent years, with the competition in the video industry, such as Netflix established streaming site is experiencing severe tests. According to NetflixQ2 results show that the company in the second quarter on revenue of $1.07 billion, compared with $889 million in the same period last year, Net profit of $29 million, $6 million sharply increased 3.8 times than the same period last year. For a video web site, although there are lots of factors affect the development of Netflix precise recommendation mechanism based on large data, as one of the powerful tools for the turnaround. wired magazine had an interview with the company’s innovation and personality algorithm division vice President Carlos Gomez – Mr Uribe, as well as engineering director Xavier Amatriain. Two people in the interview talked about Netflix, focus on how to through the upgrade algorithm, “control” viewing behavior of users.

q: how do you perform the recommended?

a: in spite of the huge differences between people, but if a careful analysis, will find us there are a lot in common. In a certain time, for example, most people will choose to watch the same type of program. Netflix can collect user viewing data, analysis of the user’s habits, and through scientific algorithm in common between users. For example, like the famous director almodovar, director of the film “all about my mother”) of users, may because of personal interests like his movies, but in our opinion, like the same director of users can be classified as category.

q: has the huge data is not equal to can do precise classification. You to the classification of different video and characteristic is how to divide?

a: there’s still mainly adopts the traditional artificial way. We have a team of more than 40 people, manual writing and add video classification label. Of course, most of these workers are self-employed. In addition, all the video introduction and rating on the Netflix, comes from the television and movie lovers, many of whom have related experience in entertainment industry. Although the artificial way there are personal bias, but we will carry on the training, and their professional quality is very high.

q: from DVD sales to focus on the online streaming media service, in this process, any change in your recommendation service?

a: we feel is more difficult. DVD sales and other business in the form of the original by mail and the actual purchasing behavior, the user can let’s take a quick and accurate understanding of the needs of its users. Now, in the era of streaming media services, optional the gender is very strong in the process of watching. A user doesn’t like the video, it may be in play after two seconds, was turned off. As a result, it is hard to like before, collect user feedback information easily. In addition, many users do not understand the importance of feedback interaction, therefore, they are too lazy to participate.

q: so, once, as the cornerstone of video site evaluation in advance, become meaningless?

a: I’m afraid so. From providing users with more accurate, fluent viewing experience, science is far more important than the evaluation of algorithm.

q: what do you mean, Netflix can track all of my watch?

a: basically. Users to watch video on Netflix, we can gather all of the user’s behavior data. Whether it is what you have seen, you search any, or what you just read, it can be recorded. Of course, these data just to used for after the budget. Through the analysis, Netflix can be for the purpose of each user’s behavior is optimized. Say specific point, the algorithm, we can according to user’s mutual things classified summary. In view of the user at the same interest, we will recommend the same type of video. With huge data sorting, Netflix can infer every user’s needs. .

q: for the same person different times the recommended content is exactly the same?

a: it doesn’t. We have joined the time factor in the algorithm. Netflix will according to the different time periods, respectively for the user to push a different video program. In addition, we also can according to different regions, the push relevant local video. Although this interactive recommended mechanisms exist larger technical problem, but we hope the company can do better in the future.

q: why Netflix can recommend me some video samsung, or a lower rating?

a: I have to clarify the fact, turnip greens, his taste, the lower ratings for every video is not you don’t like. Obviously, different user preferences, and Netflix’s recommendation mechanism is determined according to the data analysis and scientific algorithm. It will only recommend it think video to meet the needs of users, rather than high recommendation on video for the user.

q: anyway, we are to escape but Netflix data collected? !

a: ha ha, I think so. Many users told us that they often watch on Netflix is foreign movies and documentaries. In fact, according to the data shows that users watch the frequent is not those things.

q: for a video, in position to different web page, will affect the user viewing behavior?

a: prevalence location decision. We found that the more close to the top of the video, the more easily by the user. In addition, those at page at the top of the video and watch the most frequently.

q: Netflix’s recommendation mechanism is the biggest difference between with other web site?

a: Netflix and the difference from other sites, it will recommend mechanism as the basis of providing quality services. On eBay, 90% of the recommended information comes from its search behavior. And on Netflix, we will first accurate recommendations for users. Search function is recommended after the failure, the user can choose the way.

q: what do you think of the limitations of the present situation of Netflix’s recommendation algorithm is what?

a: maybe is recommended in the content is not comprehensive enough. Due to the number of video limit and the algorithm is not perfect, sometimes users may find interested in the same type of video.