Author/ Author of "Finance and Economics" Weekly Zeng Guang Xue Yongwei
Editor/ Dong Yuqing
If the "recommendation algorithm" is a Pandora's box, then this box is opened by the Internet companies themselves.
Now, a series of negative effects of the abuse of recommendation algorithms are being contained. On March 1, 2022, the Provisions on the Recommendation and Administration of Internet Information Service Algorithms jointly issued by the Ministry of Industry and Information Technology, the Cyberspace Administration of China and other departments came into effect, setting a precedent for algorithm specifications on a global scale.
As of March 28, according to incomplete statistics from Caijing Tianxia Weekly, apps such as Douyin, WeChat, Taobao, Meituan, Dianping, Today's Headlines, Baidu, Weibo, Xiaohongshu, Tencent News, Bilibili, and Kuaishou have launched the closing buttons for personalized content recommendation and personalized advertising recommendation.
This means that people can choose to say goodbye to this fear of being dominated by algorithms.
However, the current implementation of mainstream apps is not ideal. The Internet manufacturers are obviously unwilling to hand over the magic box, but have gone all out to "hide" the algorithm close button so that users cannot easily close it, so as to minimize the impact of the new algorithm regulations on the platform.
It is not only the Internet giants that are inseparable from the algorithm. From the user's point of view, it has been trained by accurate recommendation algorithms for many years, and many users are not willing to turn off personalized recommendations in order to accept "inaccurate" services.
The recommendation algorithm is in a cage, but its tentacles are still everywhere.
Algorithms are hard to quit
Fan Ting, an employee of the big factory, brushed her circle of friends boredly on a weekday afternoon. After just two swipes of my finger, I saw an advertisement for a Swedish designer brand. She has an impression of this clothing brand, because she has placed an order on Taobao twice. However, a fisherman's hat sold for 350 yuan, which was slightly more expensive, so she did not consume much.
"The advertisements pushed by the WeChat circle of friends are generally one notch higher than my daily consumption." Fan Ting said, "The most outrageous one was to brush up on the Lamborghini advertisement."
Wang Qiqi, who did product testing, just finished talking to her colleagues the day before, "Do you want the king's glory joint model to protect Shubao", and at noon the next day, she brushed the video of this product on Douyin. She exclaimed, "Am I being monitored? What makes Wang Qiqi even more confused is that sometimes it is just something that is talked about in daily chat, and the next second vibrato will be pushed to himself.
Dilling, who does front-end technology, is more sensitive to recommendation algorithms. She is an old user of Douyin, and recently because she searched for fitness-related content several times, Douyin began to recommend fitness classes to her every day. She also found that because she often shared videos with her boyfriend on Douyin, slowly, the video pool she and her boyfriend brushed became more and more convergent, "Some of the videos he brushed, I watched, is what I brushed the day before." ”
These are not individual phenomena, but the lives of the masses that have truly been altered by algorithms.
With the large-scale application of recommendation algorithms in the past decade, issues such as personal privacy, social ethics, and recommendation effects have also begun to receive attention. In fact, in August 2021, the State Internet Information Office issued the Provisions on the Administration of Algorithm Recommendation of Internet Information Services, proposing that algorithm recommendation service providers should provide users with options that do not target individual characteristics, or provide users with convenient options to turn off algorithm recommendation services. The regulation came into force on March 1, 2022.
Although the new algorithm regulations clearly stipulate that the platform must launch the algorithm close key in a prominent position, from the current cumbersome procedures for the algorithm to close the platform, the Internet giants are not willing to really give the choice to the user in the short term.
As the most well-known representative application of the domestic recommendation algorithm, it takes 7 steps for Douyin to close personalized content recommendation: open the App, click "I", click the navigation bar, click "Settings", click "General Settings", click "Manage Personalized Content Recommendation", and click the close button on the right side of "Personalized Content Recommendation". The whole process is very tedious.
In addition to the content of the interest recommendation, there is also "personalized advertising", some practitioners told the "Finance world" weekly, usually these two pieces of business in the company part of the different business lines, but reflected in the user port, sometimes the two functions of the closing key may be set together, convenient for users to manage. However, TikTok's aren't together.
"Finance and Economics" Weekly also tested Xiaohongshu, Taobao, Weibo and other platforms, and wanted to smoothly close the recommendation algorithm, and the operation steps were almost all more than 5.
Recommendation algorithm off button for different platforms. Photo/ Douyin, Little Red Book, Taobao
What is the original intention of this highly complex operation? According to this, the weekly magazine "Finance and Economics" asked the official inquiry of Douyin, and as of press time, it has not received a reply.
But what is certain is that shutting down the recommendation algorithm will have a direct impact on the commercialization of the Internet giant. According to an internal Fcaebook report in 2014, turning off the recommendation algorithm directly leads to a series of consequences such as shorter browsing time for users, a decrease in the frequency of publishing information, and a decrease in the frequency of returning and logging into the platform.
In fact, Senior Algorithm Engineer Junhao revealed to the "Finance and Economics" Weekly that at present, as long as it is an Internet manufacturer, it is now extremely dependent on the algorithm, and the hidden depth of the off key of different App algorithms also represents the degree of their dependence on the recommended algorithm to a certain extent.
"Content and community apps with large traffic will not be bad in recommending algorithms," said Wang Qiqi, who has tested many competitors. Di Lin, who has been doing technology in large factories, mentioned that for an Internet company, the search and recommendation department is often the most core department, and its recruitment requirements and staffing have maintained a high level, "Accurate recommendation and effective growth are the most important things for a company." ”
But search and recommendation are fundamentally different in business logic. Search users often have obvious purpose, need to arrive at new content according to the content typed by the user, and the recommended user may have no purpose, the platform does not necessarily need to be fully recommended according to the user's interest, but also can guide people's interest.
Of course, not all Internet companies rely on recommendation algorithms, especially some small companies and products. For these companies or products, because the amount of data is not so large that it must rely on algorithms, it is more efficient to use simple filtering.
A large manufacturer product operator told Caijing Tianxia Weekly that a list platform product she operated before did not use a recommendation algorithm, and only needed to artificially recommend some list content to the page, "This kind of product itself cannot occupy too much time for users." She also runs an information distribution platform and often chooses to use "human referrals" more often for reasons of technical cost and content professionalism.
But for head apps that measure in hundreds of millions of units of users, algorithms become essential. Especially when the amount of data in the entire industry is rising exponentially, the head Internet platform has hundreds of millions of users. Massive amounts of data and huge user scale must be filtered and improved with the help of algorithms.
Algorithms have become an indispensable infrastructure of the Internet, and have become the "honey and sugar" that the Internet cannot quit.
Ruled by the "Recommended" column
The widespread use of recommendation algorithms in China is inseparable from the popularity of mobile Internet and the rise of ByteDance. In the ten years since the mobile Internet has soared, recommendation algorithms have also gradually dominated china's Internet.
In March 2012, ByteDance was established, and the main product today's headlines were launched in August of that year. As the first news and information App in China to use machine recommendation as the main distribution method, today's headlines swept 10 million users in only 90 days after it was launched, and for the first time in China, the concept of "recommendation algorithm" was ignited.
Before today's headlines, the domestic information platforms were all editorial-centric, mainly manual distribution, and today's headlines opened a precedent for machine distribution of content.
Junhao revealed to the "Finance world" weekly that ByteDance is not the first Internet company in China to start making recommendation algorithms, later than Baidu and Ali. But with the "headlines of the day", ByteDance quickly made a name for itself and became a rising star in the recommendation algorithm, so much so that for a long time, the name of the company was called "today's headlines".
Since then, with the rise of Douyin and TikTok, Byte's strength in the field of recommendation algorithms has been further enhanced, and even become synonymous with recommendation algorithms to a certain extent.
In September 2016, Douyin 1.0 was launched, and two years later, this short video platform quickly swept across The Country and became one of the two most popular short video apps in China.
Compared with today's headlines, Douyin is more aggressive and bold in algorithmic recommendations. In Douyin, users watch short videos through repeated "under-stroke" operations, and in the process of continuous down-stroke, the algorithm continuously "learns" the user's preferences, and constantly trains the model according to the user's behavior characteristics such as likes, favorites, and attention, so that the short videos recommended by the system are more in line with the user's preferences.
Photo/Visual China
At the same time that Douyin swept the country, the international version of TikTok has also risen rapidly. In August 2017, Byte founded TikTok overseas, later acquired Musical.ly and integrated the two sides, and from 2019 onwards, TikTok quickly swept the world.
Like TikTok, TikTok relies heavily on recommendation algorithms. Among the top ten breakthrough technologies released by the Massachusetts Institute of Technology (MIT) in 2021, TokTok's recommendation algorithm is among them because it can give ordinary people's content a chance to be treated like celebrities and become popular, achieving a certain degree of content fairness.
After 2018, with the rapid rise of Douyin, algorithm recommendation has gradually become the mainstream way of domestic information distribution, and information platforms, e-commerce platforms, and grass planting platforms have bet on recommendation algorithms, and different types of Apps have been ruled by the "Recommendation" column.