How does the algorithm "know" what we put in the "Shopping cart"?

People have been accustomed to open a Web page at the same time, may be pushed to send all kinds of merchandise advertising, sometimes we will tease that they are "big data". But few people know that in order for this unobtrusive product ad to appear on the right screen, a spectacular cluster battle has been staged behind the screen:

Different advertising agencies in the system of "you" to identify and calculate, according to the identification results of each bid information sent to the ad auction platform, the ad auction platform by the price of the highest bidder, the highest bidder to obtain the opportunity to display the ad, and finally a product ad appears on the "You" screen.

This form of advertising through real-time bidding is the most common way for advertisers in the internet era to do redirection and precision, and is a form of advertising that relies entirely on "algorithms", as the above series of bidding actions must be completed within a second.

The most well-known company in the field of redirection is the French-born advertising (Beijing) Co., Ltd. (Criteo), which was born in 2005 by the "algorithm" has a market value of $2.7 billion, starting from 2010, the annual total revenue has increased by more than 60%, 2015 total revenue exceeded 1.1 billion euros.

　　Compute-driven precision delivery

Although I don't know who "you" are, but when you load a Web page, the system has "calculated" You from 2 dimensions in a very short period of time: how much it costs to be able to buy this opportunity to display your ads, and the way ads appear on your screen after a successful purchase is more likely to inspire your desire to buy.

"In the course of advertising, the engine and algorithm will have a recommendation for what ad you are going to run. For example, you see in the Internet store in the evening is a pair of red high heels, watched for a long while did not buy out, we will vote for your ad, may be some red high heels, may be a pair of boots, we would have an artificial intelligence engine to predict. ”

Cheng, president of Criteo China, explained that all the ads displayed were based on massive data and intelligent algorithms, classifying consumers ' specific behavior and willingness to buy, massive amounts of data coming from PC-side cookies, and mobile-based predictions by ID.

In a seemingly simple ad delivery process, the "algorithm" is almost ubiquitous. "Bid algorithm" to measure the value of each user in real time, give the right price to bid; "Predictive algorithm" after finding the right users, the more likely to inspire users to buy product ads, "recommended algorithm" is tailored to the user to recommend appropriate products.

For example, Criteo ads for an advertiser, after finding the right user, will first to the user to do conversion rate prediction, high conversion rate of high bid, "such as I think this person's value is 5, my bid is close to 5, if this person value is 2, I use 5 to bid, then I may lose money, Because his conversion rate is lower than expected. Cheng said, the description of only a few words, behind the need for the system has a strong computational power.

Criteo China President Cheng

In addition, Criteo also need to use the Creative optimization engine to understand and cater to the user's preferences, the same product, need to provide different ways, color depth, image representation and other programs, Criteo to prepare up to 30,000 creative combinations to attract the screen before the taste of different consumers to click.

Monetate, a well-known U.S. e-commerce technology company, reported that 97% of users will not have any purchases when they leave the site, and 90% of sales may occur within 15 days after the consumer first touches the product on the site.

Based on this, Criteo uses algorithmic engines to help advertisers identify and find potential consumers, predict what they might buy, and then push personalized ads to their eyes, motivating users to click and jump until they finally buy.

For example, a "potential consumer" that could be turned into a consumer for 15 days will be directed back to the product they browsed for the first time they visited the site and eventually to a purchase. In addition, it tracks lost users within 400 days of their last visit, analyzes them based on their browsing behavior and interests, and then recommends them to other older visitors or dormant users to interact with the site again.

　　Only fuzzy recognition, no precise portrait.

In the age of artificial intelligence, the author Jerry Kaplan a description: "At any time, when you are shopping, visiting a website or commenting, there is a mercenary of electronic agents observing you in the dark." ”

In the Internet age, the enterprise exhausted methods to obtain information to the user portrait, and strive to understand users more than users, but "user portrait" often let users have no sense of security.

Companies that do user portraits often get information about their gender, education, income level, social preferences, etc. from the Internet and social media, and then push merchandise ads based on this data. Cheng that this "user portrait" approach is not accurate compared to redirecting, because most of the time it is speculation.

Criteo's redirect ad does not care about the user's true identity, does not give the user a portrait, does not give the user a personal identity tag, or even does not care about the content of the media where the ad is located.

"We are only concerned about the price of this advertised value in the auction process, and if you buy it at a very accurate rate, you will have a good control over the cost behind the conversion." Cheng explained that Criteo used first-party data when making predictions and delivery, and that there was no way to paint a portrait of each individual user.

For example, a user has been to Dangdang browsing, as long as his cookie has not changed, although Criteo do not know who the user is, but you can know that the account browsing when the many product pages, using these first-party data is enough to do engine analysis. Criteo need not more user data, but to predict the conversion rate is high enough, the bid is accurate, which depends on machine learning ability and algorithm improvement.

And back to the beginning of the article "algorithm how to know what we put in the shopping cart?" "In fact, with the improvement of machine learning ability, advertising is more accurate, not" algorithm "know what we put in the shopping cart, but the algorithm imperceptible in the impact of our purchase.

http://news.cnblogs.com/n/550869/

How does the algorithm "know" what we put in the "Shopping cart"?