The algorithm recommendation mechanism of Toutiao and Douyin is not as "stupid" as people say on the Internet!

In 1865, on the flat land outside Yongdingmen in Beijing, the British man Durant started operating a steam locomotive on a railway only a mile long. At that time, the common people in the capital had never heard of this. According to Li Yuerui's "Spring Ice Room Wild Travel Notes" in the Qing Dynasty, "The people of the capital were shocked and shocked, thinking it was a monster. The whole country was in a frenzy and almost suffered a great change."

The final outcome was that the Infantry Commander's Office ordered it to be demolished. Similarly, the opening of the Shanghai Wusong Railway 10 years later also caused a stir, and the Qing government ended up purchasing the railway and demolishing it.

The same thing is happening to algorithms today. There is a special word in English for this phenomenon, called Technophobia. Similar to the popular culture of "Frankenstein" in the last century, people have a natural fear of new technologies.

To put it simply, an algorithm is a set of evaluation mechanisms. This mechanism is effective for all users of the platform. It improves platform functionality based on a series of user feedback, enhances user experience, and ultimately forms a virtuous and recyclable ecosystem for the platform.

But the demonization of algorithms continues. For example, recent articles like "What is more terrifying than TikTok making people depraved is Toutiao making us stupid." These articles emphasize two points: 1. The algorithm only recommends what you like to see. 2. Platforms such as Toutiao rely solely on algorithms.

These arguments intentionally simplify objective facts and are actually very standard Technophobia. When artificial intelligence is used for content distribution, the ensuing discussion on how to effectively overlap user needs with recommended content and clearly define their boundaries will be a very long-term topic that cannot be concluded simply and crudely.

-01-

The content mechanisms of Toutiao and Douyin have been evolving

The author of the article "Becoming Stupid" may not have used the current Toutiao and Douyin, and his knowledge is still stuck in the past. Internet comments are like this. There are always people who like to look at and judge others with a fixed perspective, which leads to wrong conclusions.

Toutiao is well aware that its algorithm is imperfect and has its own limitations. Therefore, it needs another distribution method to make up for the shortcomings and drawbacks of the algorithm, or to find the second wheel to drive the Toutiao system. Now it seems that the second wheel is fan distribution, or social distribution. Today's Toutiao has updated seven major versions. It is no longer a pure algorithm, but a combination of "algorithm + social + search + question and answer".

For example, you can use the search function to spontaneously find any information that is valuable to you. If you like philosophy, you can search for keywords such as "Nietzsche", "Plato", "Foucault" on Toutiao to get relevant information and communicate with people. In fact, "actively searching" and obtaining information is what all platforms hope everyone will do, because this itself can create more precise value.

The value of the algorithm also lies in long-tail distribution. Even niche information can get good recommendations. At the same time, if the information is rich enough, people with niche interests can also find valuable information. For example, some less popular movies, old photos from the 1980s, and university campus information. These information that were originally scattered in various corners of the Internet are collected into a recommendation box and appear in front of you, instead of being buried in the ocean of information as in the past.

And if you look at the entire homepage of Toutiao, it actually has four sections: the first is the search area, where users can actively search for information they are interested in; the second is the pinned area, which is about national affairs; the third is the headlines area, which is about important media news, usually labeled "hot"; the fourth is the personalized area, which is about information that users prefer, including both information that users subscribe to on their own and information recommended by the system.

In other words, in the actual product mechanism of Toutiao, the algorithm does not just push content that interests you, nor is there a situation where you can only see content recommended by the algorithm. The success of Toutiao mainly relies on its understanding of traffic and the execution capabilities formed by the company's operations, algorithms, and data-driven thinking. Algorithms are important, but they are only one part of the equation.

-02-

The algorithm does not just recommend what you like.

If you think that the algorithm will recommend you what you like, then you are oversimplifying the algorithm. In fact, this statement has existed for a long time. In communication studies, it is called the so-called echo chamber situation. Many people are worried that the Internet's content recommendation algorithm will aggravate this so-called echo chamber effect - what you like will be recommended, and in the end, all the content will be the same type, and you won't be able to see anything else. But it turns out that this didn't happen, and no Internet company would be stupid enough to do so.

The reason is simple. Human nature is to get tired of the old and like the new. The pursuit of diversity is a very basic need. It is inhumane to always push whatever you like. It's like giving you the same food every meal without showing you the menu. No restaurant would do that.

For example, let's take TikTok as an experiment. Use a new mobile phone to open TikTok without logging in and randomly watch 10 videos. The statistics are as follows: daily life (2 videos), simple drawings (1 video), cute pets (1 video), outdoor documentaries (1 video), young ladies (1 video), English teaching (1 video), family and parenting (1 video), fitness (1 video), and cooking (1 video).

I liked 50 videos of young ladies in a row while logged in. After restarting, the next 50 videos recommended by Douyin were: 24 of young ladies, 9 of daily life, 6 of travel scenery, 5 of dancing, 3 of positive energy, 1 of cute pets, 1 of science, and 1 of fitness. In other words, even if you are crazy about the girl, Douyin will not be able to push her videos to you continuously.

In fact, the algorithm has a certain breaking-up mechanism based on user interest preferences, so there will not be a phenomenon of continuous identical recommendations. Thinking that algorithms simply record your single preference and then make recommendations based on that preference is, first of all, a disdain for human nature, and secondly, a disdain for algorithms. A person with a sound personality can never have only one hobby - liking young ladies does not prevent you from also liking cute pets, scenery from all over the world, music and dance.

Moreover, due to the law of diminishing marginal utility, continuously pushing the same topic is a very unprofitable and thankless task.

There is also "Weber's Law" in social psychology. When a person experiences strong stimulation, further stimulation becomes insignificant. In terms of psychological feelings, the first stimulation will dilute the second stimulation.

This also explains why people always "love new things and dislike old things". A person's interests and hobbies cannot always remain the same.

It can be said that even from the perspective of meeting user needs, it is impossible for the platform to always recommend young ladies to you.

-03-

The algorithm itself has the ability to explore

What's more, algorithms are not mechanical arithmetic.

If this is the case, major technology companies will not have to fight so hard to compete for algorithm engineers who graduated from prestigious universities. Algorithm engineers are not worthy of a million-dollar annual salary. Google, Microsoft, etc. will not have to hire a large number of high-end talents to optimize recommendation algorithms. The charm of algorithms lies in continuous learning, iteration and evolution. The reason why the robot can defeat the world's top chess players is because of its powerful learning ability, which forms an extremely high competitive barrier. Exploratory nature is one of the inherent characteristics of recommendation algorithms.

This involves a knowledge point: collaborative recommendation.

The collaborative recommendation of the algorithm, in addition to the content itself, also includes user-based recommendations, which expands the exploration capabilities of recommendations by analyzing the similarities between different users. Zhang San likes technology, finance, and sports, and Li Si likes technology, finance, sports, and health content, so the algorithm will try to push health information to Zhang San. Because in the eyes of the recommendation system, Zhang San and Li Si are similar people. As a result, Zhang San can receive health information that was originally not within his interest range.

The recommendation system recommends content that a group of people similar to you may be interested in. The content they are interested in is likely to be what you will be interested in, but you didn’t know it before, the so-called “you don’t know that you don’t know”. This enriches the content to a certain extent.

-04-

Algorithms are easier to break through circles than social networks

In comparison, the circle of friends has a greater "echo wall" effect, and people often fall into it without realizing it. Because all the information in your circle of friends is completely customized by you, and because the content producers are completely screened by yourself, people cannot hear views and opinions that are different from their own, and it is easy to form prejudices against the same type of people or the same group of people. Recommending content through algorithms rather than based on opinion can ensure a greater variety of viewpoints.

If short video apps had not emerged, workers and ordinary people at the bottom of society would continue to be ignored by this society.

These people could be the guy who delivers your takeaway every day, or they could be the apprentice who cuts your hair, or they could be the workers on the assembly line that makes the phone you are holding in your hand. They have never been on the hot searches on Weibo, never visit Zhihu, and will never appear in your WeChat Moments. It is the algorithm that brings together people from all over the country and of all identities without distinction into your mobile phone. Only in this way can you see the rich and diverse aspects of life through the mobile phone screen.

For example, Tik Tok will even re-recommend “high-quality old content” in its database to give it more exposure. The reason why these old works can be "ignited" is generally because these accounts have published enough vertical content, the labels have become clearer, and the recommendation algorithm can help match these high-quality content to more accurate users. On platforms like WeChat official accounts, it is rare for high-quality content from a month ago to be "dug up" and become popular again, which is actually a waste of massive high-quality content.

A high-quality recommendation system is essentially a process of information noise reduction. From a huge content pool, we select valuable information for you and recommend it to you, reducing your cost of searching, screening and organizing information.

After all, algorithms and apps are just tools. Indeed, algorithms as tools should be continuously improved. But how to use it depends on oneself.

Russell once put forward a point of view: No matter what you are studying or thinking about any point of view, please just ask yourself "what are the facts" and "what is the truth confirmed by these facts". Never allow yourself to be influenced by what you would rather believe, or what you think would be more beneficial to society if people believed it. Just simply examine what are the facts.

What makes us stupid is not the algorithm, but the loss of rational scrutiny.

Source: Fat Cat's Tavern