In-depth analysis of Tik Tok’s recommendation algorithm!

As the saying goes, only by knowing yourself and your enemy can you win every battle.

As a short video operator, the first step is to have a thorough understanding of Douyin’s user profile, content features, and recommendation mechanism. Otherwise, if you start blindly, the biggest result may be a hasty ending.

Tik Tok recommendation algorithm

What is the TikTok recommendation algorithm?

To put it simply, an algorithm is a set of evaluation mechanisms. This mechanism is effective for all users of the platform, whether they are content producers (people who shoot videos) or content consumers (people who watch videos). In many cases, we are both producers and consumers. Tik Tok recommendation algorithm

Every action we take on the platform is like a clear instruction, and the platform judges our nature based on these instructions. Tik Tok recommendation algorithm

We will be classified into high-quality users, silent users, lost users, recoverable users, etc.; they will also determine whether we are a marketing account and whether we have any illegal operations. Tik Tok recommendation algorithm

If our account is a marketing one that does not conform to the tone of Douyin, the platform will put our account in a small black room; on the contrary, if it is judged that we are a high-quality user, the platform will give us certain support.

What is the use of algorithms?

The greatest use of algorithms for platforms is to manage user data on their own platforms, and to improve platform functions based on a series of user feedback behaviors, thereby improving user experience and allowing the platform to attract and retain more users, ultimately allowing the platform to form a virtuous and recyclable ecosystem.

What are the benefits of algorithms?

The benefits of algorithms to content producers: Since we want to attract fans on other people’s turf, we must understand their rules. Tik Tok recommendation algorithm

Just like pursuing a girl, you have to understand her likes and dislikes, only then will you have a chance to find a way into her heart. What's more, it is much easier to understand the platform than to understand girls! Tik Tok recommendation algorithm

As long as we are aware of the platform's recommendation mechanism, we can consciously design our own behavior to guide the platform to determine that we are high-quality users and allocate us more, more accurate traffic and higher permissions.

The benefits of algorithms to content consumers: After you browse Taobao, Douyin, or Toutiao a lot, when you open the app next time, do you feel that many of the recommended content are what you like and are more interested in reading?

In fact, the platform improves user experience in order to retain you. It will analyze your interests based on your behavior, then label you and recommend content from content producers with similar labels to you. You are then people in the same pool.

The process of the Douyin recommendation algorithm can basically be summarized by the following figure:

Step 1: Review

After uploading the video, the first step is machine review.

Then the review includes the video images, title keywords, and video text; for example, whether there are advertisements, whether there are watermarks or logos, whether the content is nude, indecent, bloody, etc. If there is content prohibited by the platform, our video will be rejected or restricted (only you can see the content you posted).

If there are any violations, the system may enter the manual review stage early.

Step 2: Intelligent Distribution

If there are no keyword violations or image problems, the system will combine the keywords to match around 200 to 300 users, which is what we call the initial traffic pool.

After the video is released, the system will add labels to your video based on its content (such as travel, beauty, food, Chongqing, Xi'an, island...), and then the machine will recommend it to a small number of people who may be interested in the labels of your video, and calculate the number of comments, likes and shares of the audience within a unit time.

Tik Tok recommendation algorithm

The specific formula is: Popularity = A×video completion rate + B×number of comments + C×number of likes + D×number of shares . The coefficients A, B, C, and D will be fine-tuned in real time according to the overall algorithm. Generally speaking: playback volume (completion rate)>likes>comments>reposts. This is the first recommendation.

Step 3: Expand referrals

If your video receives good audience feedback after the first recommendation, then your video will be recommended to more potential viewers, which we call expanded recommendation.

The mechanism is the same as the first recommendation, and the number of viewers reached this time is approximately 1,000-5,000 people.

If the feedback from the second recommendation is good, the platform will recommend it for the third time. The third time will get tens of thousands or hundreds of thousands of traffic, and so on. If the feedback is still good, the platform will use big data algorithms combined with manual review mechanisms to measure whether your content can become popular.

According to the practical experience of DouShang Commune, a video can basically become popular if the video playback volume reaches more than 5,000 within 1 hour of its release, and the number of likes is higher than 100 and the number of comments is higher than 10.

Therefore, please remember the following series of numbers: 1-5000-100-10;

What does it mean? That is to say, it is best if the video you publish can be played over 5,000 times within 1 hour, and the number of likes can be greater than 100, and the number of comments can be greater than 10; then, the chance of being recommended by the system will be much greater, and it will basically be close to becoming popular.

Potential for a hit video

I believe that when you operate an account, you often feel expectant or even uneasy. That is, after the video is released, I will check the number of views from time to time to see if the video will become a hit. Tik Tok recommendation algorithm

According to the practice of DouShang Commune, we actually pay more attention to the first hour after the work is published. Generally speaking, if the system provides more than 5,000 traffic and more than 50 likes in the first hour, with a broadcast-to-like ratio of 100:1, there is a 90% probability that it will enter the next larger traffic pool.

If the user interaction in the next traffic pool exceeds 100:1, it may continue to explode, and generally 10 million views will become a bottleneck (after the video becomes popular, it will be manually reviewed again, and if the human resources feel that the content is skirting the rules, the traffic will be directly withdrawn).

This is why many people are confused as to why the system suddenly stops recommending content when the traffic is clearly running well. Tik Tok recommendation algorithm

In fact, sometimes even if the content is good, it may not be liked by the system. The decision still lies with the platform. Especially those involving politics, privacy, exposure, and sensitive factors such as mutual following are likely to be stopped or even directly blocked.

But if you want to enter the million pool and have a small explosion, the likes rate must be maintained at around 10% in the first hour. Otherwise, no hope. To enter the 10 million pool, you need a higher likes rate. Tik Tok recommendation algorithm

ps. Regarding account violations, the requirements and restrictions of the Douyin platform will only increase. After all, the daily active users are there. If there is no strict control, Douyin may become the next connotation joke.

3Why 100w views?

Need to reach 10% like rate?

According to our actual operation and feedback from members of Dou Shang Commune, 3% is the average rate of likes on Douyin, so if your work does not get a 3% likes rate in the first hour, there is basically no possibility of it becoming a big hit.

For a work to impress your fans, it needs a at least 3% like rate. Tik Tok recommendation algorithm

When a work is pushed to strangers' mobile phones by tens of thousands of broadcasts, if there is no 10% likes rate in the early stage, it is very difficult to deal with the problem of how many likes strangers can give you after tens of thousands of broadcasts. Tik Tok recommendation algorithm

For a work with a 10% likes rate in the early stage, the likes rate will drop sharply after entering the big pool, because your work is vertical, fans like it, but strangers don’t like it very much.

After giving you tens of thousands of broadcasts, the likes rate will drop to about 7%. This is still a very good playback rate, which can continue to support your work to roll into the 100,000-level pool, or even the 1 million pool.

After entering the million-level pool, a larger wave of people will come, so your broadcast-like rate will further drop to around 5%, or even around 3.5%. That’s why your work feels like it’s flamed out. Because it has reached the average level of Tik Tok. Tik Tok recommendation algorithm

Therefore, after a work is released, one can basically predict whether it will become popular within one or two hours, or about 30 minutes. The examples of overnight popularity or future popularity belong to another category - that is, what we call secondary recommendation and digging up graves; the applicable algorithms are different. Tik Tok recommendation algorithm

What is a secondary referral?

The so-called secondary recommendation is commonly known as "digging graves". In fact, it comes from a closed-loop effect.

For example, if after a work of yours is released, Douyin actively pushes 2,000 views and gets 0 likes, it must have been a failure. But if there is a closed-loop effect, someone comes in through another video and likes this work. If 200 people can come in and like this work, it will get 200 likes. At this time, the likes rate of this work has reached 10%.

Tik Tok is scanning at all times and will restart the push of this work. This is called grave digging. There is just an algorithm behind it. Second recommendation and digging up old pasts all mean the same thing, and this is the principle behind it.

Author: Peng Qian

Source: 36Kr