Douyin recommendation review mechanism, see here!

The number of monthly active users of Douyin worldwide has exceeded 500 million. In the era of "two Weibo and one Douyin", this is a traffic pool that no one wants to miss. For operators, the most important concern is the distribution mechanism of Douyin and how to cater to it and obtain more playback volume.

Douyin's distribution mechanism is a replica of Toutiao. Generally speaking, the order of priority is: deduplication mechanism, review mechanism, feature recognition, recommendation mechanism and manual intervention.

Let’s analyze them one by one.

1. Deduplication Mechanism

Eliminate duplicate videos. If your video has been posted by someone else, the possibility of it being recommended will be much lower. So originality is very important.

The duplicate videos here include other people’s videos and highly similar videos.

If you take the photo yourself with your phone, it is usually original. Generally, deduplication is performed on uploaded videos so that they do not appear simultaneously or repeatedly in the user's video stream.

Of course, Douyin’s deduplication mechanism is not as strict as Toutiao’s, otherwise all those imitation videos might have failed.

II. Audit Mechanism

Like Toutiao, Douyin’s review is also divided into machine review and manual review.

Generally, machine review is the main method, and some reviews that machines cannot judge are done manually.

The review generally involves reviewing the video content and video description/title. The main review is to see whether there is sensitive information and whether there is a QR code/phone number/link.

There is an interception library in the machine review algorithm. After the video is released, it enters the review state immediately. The machine automatically compares the video title and content to see if they match the library. If they match, it will not pass the review.

Therefore, it is a good idea to post more videos with positive energy and main themes. Of course, if you want to send promotional information, then a blue v certification is still necessary. Usually you can get certification quickly by finding a service provider.

3. Feature Recognition

After the video passes the review, the Tik Tok system will label the published video according to the content and title of the video, match it with the relevant user groups, and prepare to push the video to this group of people.

IV. Recommendation Mechanism

All the preparations are done, and now we are ready to start pushing.

Considering that the tags identified through features may not completely match the interests of the users to be pushed, the system adopts a phased push to the user group in batches.

Generally speaking, push notifications are sent to a small number of users (such as 100 people) first. If 10% of them interact, it means that the recommendation is probably accurate, and the system will automatically expand the push scope (such as 1,000 people). If there is still more than 10% interaction, the push volume will continue to increase.

The interaction here refers to the playback completion rate, likes, comments, and sharing.

By analogy, the recommendation mechanism can be understood as a stone thrown into a body of water, which creates a small wave at first and then slowly expands outwards.

5. Human Intervention

Video content review is much more difficult than text review, and machines cannot do everything.

Since the machine cannot accurately determine whether a video is illegal or what its quality is, and since the machine makes judgments based on past data, it cannot completely predict how users post videos. After considering many factors, the artificial intervention mechanism was introduced.

Douyin is also recruiting a large number of video content reviewers, which is the role of the human intervention mechanism. The reviewers' experience can be used to determine whether the published videos are in violation of regulations and report them, which can fill some of the loopholes in Douyin's machine review.

Finally, let me tell you a little trick.

The review of Douyin is the same as that of other videos. It restores the video into frame-by-frame images and then identifies whether there is any illegal content in the images. The recognition accuracy is as high as 99.5%, so don’t be lucky, just create content honestly and then open a shopping cart, this is more realistic. (You can ignore those just for brand promotion, that is another idea)

author: network

Source: Internet