Tiger Olfactory Note: This article from the Love Fan Son network, tiger sniffing carried out related edits. in the age of explosion, intelligent recommendation is born. Content providers can make more personalized recommendations by making more detailed analysis of the user's preferences. Alexis C. Madrigal of the Atlantic website reveals the truth behind Netflix's film and television recommendations through in-depth analysis of the Netflix film Classification and interview with the vice president of Netflix products.
interpretation of the outside world: grammarWhen Netflix recommends a movie, it gives a very granular type, or a "micro-type". Alexis wanted to sort out a complete list of micro-types. Due to the outrageous number of Netflix's micro-types, he had to use automatic processing software. After a day of continuous computer operation, he saw the number of Netflix micro-types: 76,897. These micro-types are composed of some basic elements, the basic syntax is this: region + adjective + type (noun) + based on ... For background + from ... + about ... + from X-years to Y-years (region + adjectives + Noun Genre + Based on ... + from the ... + about ... + for ages x to Y) Yes, of course. , some micro-types do not conform to the above syntax, such as "have a strong heroine" and "suitable for the incurable romantic." At the suggestion of colleague Ian Bogost, Alexis also wrote a type builder to try to understand what Netflix's algorithm looks like. However, he still felt that he did not see the whole picture. So he contacted Netflix and met the company's vice president of product, Todd Yellin, who created the Netflix referral system.
Internal solutions: content tearing, quantum theoryIn an interview, Todd Yellin said he had long been waiting for people to ask the question. It was his idea to tag all the films, and he wrote a 24-page document describing the tagging method. He tagged some of the early films and dominated the whole system creation process. "My first task was: RIP the contents apart!" "he said. In 2006, Yellin organized more than 10 engineers and spent months writing a document called "Netflix quantum Theory". At the time, he called the basic elements of the film "Quantum." Today, he prefers the word "micro-tagging", because now it seems that "quantum theory" is a pretentious name. This document describes how to label various aspects of the film, involving the end of the film, the "social acceptability" of the protagonist, and so on. Labels are scalar, from 1 to 5, which means that all films have a "romantic" identity, not just those that indicate "romantic movies." Every movie has its own score, from happiness to sadness. Each episode has a label. The protagonist's work has a label. The location of the film is labeled. Netflix's engineers created a syntax based on "micro-tagging" to generate various types. This is a combination of manual and machine algorithms. Netflix's approach is somewhat similar to the Pandora's music fund program, but Netflix doesn't just show you what you like, it can also be described in words. These tag data are not only used to generate types, but also to increase the recommended personalization level. If you like action-adventure movies and have a high degree of romance, Netflix can recommend it to you. "We'll mark the romance in the film. We won't tell you how romantic it is, but we'll recommend it, "Yellin said," You'll see the action scene, how romantic it is, it depends on what we know about you. " ”
Philosophical thinking of algorithms: Mistakes are also featuresBefore the interview was over, Alexis presented an interesting chart to Yellin. Among the most popular performers on Netflix, the first was the Raymond Burr,50 ERA TV show.
Perry MasonStarring, and the seventh place is Barbara Hale, the secretary who starred in the play. Why are these two people able to surpass Meryl Streep, Samuel L. Jackson, Nicolas Cage and Sean Connery? This is a puzzle. Even Yellin do not understand why their names appear in so many "micro-types". This is a problem that cannot be explained by logic. "Let's do some philosophical thinking." In human society, life becomes interesting by accident. "When you add complex things to the world of machines, you're adding surprises that you can't imagine," he says.
Perry MasonIt will happen. Ghosts in these machines are often a byproduct of complexity. Sometimes we call it a mistake, sometimes we call it a feature. ”
Article for the author independent opinion, does not represent the Tiger sniffing network stand
Transferred from: http://www.huxiu.com/article/25983/1.html
