The secrets of Xigua Video’s recommendation mechanism revealed!

To understand the recommendation mechanism, we must first understand the audience. Each viewer's viewing interests are very different, and what the personalized recommendation mechanism does is to let each user see content that may be of interest. This kind of precise recommendation is based on the premise that the machine has a full understanding of each user. In the machine, each user is actually made up of a large amount of data, and the user's viewing interests are hidden in these data. Different data has different weights in calculating user interest, including:

• Gender, age, city;
• Accounts followed, channels frequently visited, topics of interest;
• Categories and keywords of videos watched;
• Other video types that other users of similar genres also like to watch;
• ….

The essence of the recommendation system is to match a small amount of content of interest to the audience from a massive content pool. In order to provide users with content they like or understand their needs, the platform has many angles to portray a user's profile, such as age, gender, historically browsed articles, environmental characteristics, etc. At the same time, it uses advanced AI technology to classify content. Next, the recommendation mechanism acts like a "bridge" that connects the audience and the content, pushing content to interested users continuously. This "bridge" has two characteristics:

• Interest matching: The viewer’s viewing type has the highest overlap with the content classification, and the system determines that he is most likely to be interested in the content.
• Recommendation in batches: The content will first be recommended to a group of viewers who are most likely to be interested in it. The data generated by this group of viewers will play an important role in the next recommendation of the content.

A good way to increase referral volume

Note: The following examples are only used to illustrate the impact of click-through rate and other factors on video recommendations, and do not represent actual recommendations.

For the first recommendation of a video, if the click-through rate is low and the number of reposts, comments and likes is not high, the system will consider the video unsuitable for recommendation to more users and will reduce the number of secondary recommendations. If the click-through rate and other data are high, the system will consider the video popular with users and will further increase the number of recommendations. Similarly, the number of new video recommendations is based on data such as the click-through rate of the previous recommendation.

For example, a video is recommended to 1,000 users for the first time. If the click-through rate, completion rate, etc. of these users are high, the system will determine that the users like the video very much and will recommend it to 10,000 users. If the click-through rate, completion rate, etc. of this round of recommended users remain at a high level, the system will recommend the video to 30,000 users, 50,000 users, 100,000 users... and the number of recommendations and views will continue to rise like a snowball. The recommendation volume of a new round of recommendations will gradually decrease only after the video has expired for 24 hours.

Because of this mechanism of expanding recommendations, if authors want to get more playbacks, they must work hard to maintain reading data in various dimensions (click-through rate, user playback time, number of collections, number of comments, number of reposts, etc.) at a high level.

This requires the video to:

• The title and cover image are attractive enough and clear in meaning to increase the click-through rate;
• The video content is high-quality, with excellent editing and commentary, which increases the user's playback time and playback completion rate;
• The content is detailed, giving viewers a sense of fulfillment, increasing the number of collections and user playback time;
• The viewpoints are clear, sparking discussion among the audience and increasing the number of comments and reposts.

Among them, the most important are of course the click-through rate, completion rate, reposts, comments and likes, etc. Therefore, the importance of the title and cover image is self-evident. This is also the key factor for achieving good dissemination of broadcasts in today's era of scarce attention.

Attractive titles can bring more clicks, but this does not mean that authors have to become clickbaiters who use exaggerated titles. On the contrary, clickbait titles will be identified and suppressed by the platform through technical means, limiting the number of recommendations. In addition to exaggerated titles, intensive user reports and too many negative comments are factors that limit the amount of playback recommendations. Ultimately, whether a work can get more recommendations depends on the quality of the content. Good content can bring long-term growth in traffic.

Check the work

From the above interpretation of the recommendation mechanism, it is not difficult for us to classify the main reasons for poor recommendations. First, you can check whether the work has any low-quality behavior that violates the posting standards or review standards, such as:

• Advertising information: content containing WeChat public account/Weibo/telephone contact information, etc.;
• Deduplication mechanism: There is already highly similar content on the site;
• Title: clickbait/exaggerated title/inconsistent with the content, etc.
• Low-quality content: vertical screen/black-bordered videos/horror content/vulgarity/curiosity, etc.;
• There are frequent user reports, too many negative comments, and many invalid and abnormal clicks;

"Toutiao Backstage" - " Xigua Video " - "Content Management" currently supports the function of explaining the reasons why works are not recommended well. It is very important reference information. You can read it carefully to understand it and check whether there are any problems with the work.

Common system display reasons (selected parts)

Interpretation

The work is suspected to contain vulgar content (such as: x minutes x seconds, etc.), it is recommended that you publish positive and healthy content

If this prompt appears, there may be pornographic or vulgar content in the content, title or cover. You can check each one to see if there is any problem.

The work is suspected to contain sensitive topics (such as: x minutes x seconds, etc.), it is recommended that you publish positive and healthy content

If this prompt appears, it may be that the content, title or cover contains sensitive topics. You can check each one to see if there are any problems.

The work is suspected to contain unverified information (such as: x minutes x seconds, etc.), it is recommended that you publish objective and true content

The appearance of this prompt means that the video content may quote unverified or unofficial information. It is recommended that you clearly indicate the source when quoting information, or choose objective and true content to publish.

The title of the work may be exaggerated. It is recommended that you change it to an objective and true title as soon as possible.

The appearance of this prompt indicates that the video title uses exaggerated expressions, such as "shocking", "explosive", "going viral", "scared half to death", etc. For specific examples, please refer to the title creation standards. The work can be modified within 14 days after it is published. If there is no problem with the modified title, the system will resume normal recommendation.

The title of the work does not express complete information. It is recommended that you change it to a complete title.

The appearance of this prompt indicates that the video title is incomplete or hides key information, such as "It turns out to be...", "But...", "However...", etc., which leave the sentence half-finished, use ellipsis to replace key information, or use "inside story", "revealing", "truth" and other words to replace key information. Please refer to xxxx for specific examples. The work can be modified within 14 days after it is published. If there is no problem with the modified title, the system will resume normal recommendation.

The title of the work may easily mislead people into thinking it is a real event. We suggest that you modify the title and explain the name of the material.

The appearance of this prompt indicates that the video title does not clearly explain the confusing elements, or is missing the main information, does not indicate the subject type or work name, etc. For specific examples, please refer to the title creation specifications. The work can be modified within 14 days after it is published. If there is no problem with the modified title, the system will resume normal recommendation.

The work involves outdated information (such as: x minutes x seconds, etc.), it is recommended that you choose recent and fresh topics for your creation

The appearance of this prompt indicates that the video content uses outdated news or outdated information, which affects the viewing experience. You can check whether problematic materials are used. It is recommended to choose fresh themes for creation.

The work contains advertising information or other platform/tool ​​logos (such as: x minutes x seconds, etc.). It is recommended that you avoid similar issues in subsequent works.

If this prompt appears, it means that there are other platform/tool ​​logos in the video content, title or cover, such as Weibo, WeChat, QR code, telephone contact information or other third-party platform watermarks, or other third-party editing tool watermarks or opening and ending screens. You can check each one to see if there is a problem.

The work has problems such as screen scaling or subtitle truncation (such as: x minutes x seconds, etc.). It is recommended that you avoid similar problems in subsequent works.

The appearance of this prompt indicates that the video content has problems such as the character image being truncated or distorted due to zooming in or the subtitles being incomplete, which affects the viewing experience. You can avoid such problems during post-editing.

The work involves a sudden interruption at the end or the sound and picture are out of sync (such as: x minutes x seconds, etc.). It is recommended that you ensure the content is complete in subsequent works.

The appearance of this prompt indicates that there is a sudden interruption at the end of the video content, such as the person has not finished speaking or the picture has ended, which affects the viewing experience. You can pay attention to the completeness of the content during post-editing.

The work is suspected to be too similar to other works on the platform. It is recommended that you publish high-quality, original content

The appearance of this prompt indicates that the video content is highly similar to other works already published on the platform, including content published by others and by yourself. The platform encourages and supports high-quality original content. If the work is indeed original, you can consult Xiguajun for help.

It is recommended that you make some adjustments to the problems in your work based on the above reasons and explanations to improve the overall quality of your work. Currently, the platform supports up to 5 title or cover modifications for published works within 14 days. If you encounter problems with the title or cover, timely adjustments can restore normal recommendations. If you confirm that there is no problem with the work after self-inspection, you can click the "Appeal" button next to the work in content management to submit it for review.

When the work does not violate regulations or is of low quality, but the number of recommendations is still not ideal, you need to consider whether there is room for improvement in content creation, such as whether the work has a low click-through rate:

• The verticality of the account content is low and it is not recommended to the corresponding users in a timely manner;
• Problems with the video content itself lead to low click-through rates;
• The title is bland and does not attract readers to click;
• The pictures are not attractive and not relevant to the content, or the first three pictures are of poor quality;
• The content is too low-quality, causing readers to be disgusted or even complain;

Or is the selected subject matter too boring or unpopular, and not in line with the interests of platform users, such as:

• The topic is too unpopular or obscure;
• The fields involved are too professional, obscure and difficult to understand, and have a certain distance from the mainstream masses;
• The supply of content exceeds the demand.

Author: AChasing the Wind - Self-media

Source: AChasing the Wind