A look at the logic of Xiaohongshu's homepage recommendation algorithm from the perspective of popular notes

As an operator, I have always been paying attention to Xiaohongshu . Recently, I have officially started operating Xiaohongshu and plan to become a Xiaohongshu content creator. After two weeks of operation, the number of followers increased by more than 3,500, and it received 92,000 likes and collections, and produced several popular notes with more than 100,000 readings. So today, combining the operational knowledge and the practice of popular notes, let’s talk about the logic of Xiaohongshu’s recommendation algorithm .

1. Two major characteristics of Xiaohongshu content community

1. The homepage is a double-column information flow

Because of the double-column information flow, users actively choose to open a note, so there is click-through rate (clicks/exposures) data. Click-through rate is an important factor for the system to judge the quality of note content, so the first picture of the note becomes very important. So what kind of first image in a notebook will have a high click-through rate? One is that the picture is particularly beautiful, exquisite and fancy (in line with the aesthetic orientation of the mainstream users of Xiaohongshu), and the other is that there is clear text in the picture to tell users what the content is (commonly known as a big-character poster), and it is best to have substantial content.

For products like double-column information flows, allowing users to quickly access note content is a rigid requirement, so big-character posters and clickbait headlines are inevitable. "Big-character poster" means that there is clear text in the first picture to allow users to immediately obtain a summary of the note information and make click choices quickly. There is no need to say much about the clickbait headlines, such as "must-see, quick success, lose ten pounds in five days, become white in a week, a hundred-dollar good thing" and so on.

Obviously, a large number of "big-character posters" and excessive clickbait headlines are not healthy for the community ecology. They are not real enough and not beautiful enough. Users will gradually think that this is a platform with a high threshold for producing content, reducing their desire to share ordinary daily life. Therefore, Xiaohongshu has also restricted the flow of some of this type of content. The picture below shows a search on Xiaohongshu using the keyword "learning". It can be seen that the content of the first picture of the big-character poster is very popular.

2. A high proportion of users consume content through search

Compared with Tik Tok and Kuaishou, Xiaohongshu users prefer searching. A large number of users use Xiaohongshu with specific goals (such as checking out various tutorials and promoting products), rather than just browsing and entertaining like Douyin. The signal for content creators is: choose the right keywords and discover the keywords that users like to search for. This makes it easier to gain continuous exposure in the keyword search results channel.

It is precisely because users like to consume content through search that the long tail of Xiaohongshu note content is very long. A popular content can be liked at a high frequency for a month and liked at a low frequency for half a year. Compared to the sudden rise of TikTok, the number of likes for Xiaohongshu’s popular notes is climbing relatively slowly, but the life cycle of the content is very long. According to Qinggua Media’s analysis of some KOLs’ fans, more than 50% of Xiaohongshu’s fans’ conversion rate (clicks/follows/collections) comes from Xiaohongshu searches. Regardless of whether the data is accurate or not, the traffic coming from search behavior on Xiaohongshu is indeed very large.

This feature stems from users' cognition and behavioral habits of Xiaohongshu, and is therefore almost unaffected by algorithm adjustments and operational interventions. The traffic brought by search behavior will exist for a long time.

2. Main exposure channels of content on Xiaohongshu

On the Xiaohongshu website, the main exposure of notes is based on three logics: attention logic, recommendation logic and search exposure logic. The logic of following is that people who follow you can see your content in the "Follow" tab and "Discover" tab. The recommendation logic is that the system recommends notes to people who may be interested based on their characteristics. Keywords, geographic locations and other information extracted from notes are important features of notes. Among the following five main exposure channels, 1 and 2 belong to the attention logic, 3 and 4 belong to the recommendation logic, and 5 belongs to the search exposure logic. For popular notes, 3 is the biggest exposure channel.

1. Home page - Follow tab

People who follow you will see your notes here.

2. Home page - Discovery tab

Here is a section of content posted by “people followed”, and the algorithm for this section is related to the closeness of the relationship between the user and the blogger.

3. Home page - Discovery tab

Recommended to interested users. The system will recommend other similar content to users based on their existing content browsing behavior. This traffic is relatively the largest, and most users are accustomed to browsing the homepage. Factors that affect the amount of note recommendations are generally divided into two categories: account and content. The account is the basis. Only accounts that do not violate regulations will be recommended. I will not go into details about this. Regarding content, the system has an algorithm for calculating the quality of note content. The click-through rate (clicks/exposure) and interaction (likes, favorites, comments, and reposts) of the note are the main weights. The system will decide whether to continue distributing the content based on the content quality score.

4. Home page - Nearby tab

Recommended to users within 30km.

5. Search Results Page

If the keywords of the note are set well, the note can be seen when the user searches for the keyword. If there are too many notes under a keyword, the notes with relatively high interaction data will generally receive more exposure.

There are also some relatively small probability channels:

(1) Editor’s recommendation: Recommended by the official operating account, usually because of participating in the topics that the platform is promoting.

(2) Popular notes under a topic: Notes under a certain topic are ranked high and gain traffic when users click on the topic.

3. Xiaohongshu homepage content recommendation logic

1. Recommendations based on content similarity

The feed flow on Xiaohongshu’s homepage also mainly makes recommendations based on the similarity of content, that is: if you like content A, content B with similar tags as content A will be recommended. In the recommendation system, the content tags you like are strongly correlated with your previous browsing history. The system will recommend similar content to the content you have browsed or liked (the main weight should be likes and collections). Xiaohongshu's existing recommendation algorithm provides real-time feedback and recommendations on users' favorite content tags. That is, if you like a note about breakfast, related content will appear immediately.

The format of video and text also affects the recommendation, because the related recommendations of video notes are all video notes, and text and picture notes with similar keywords will not be recommended. Like a few more videos, and you will find that your Xiaohongshu homepage will be full of videos in a short time.

2. How does the system determine whether two contents are similar?

So, how does the system determine whether two contents are similar? Mainly based on content tag similarity and image similarity.

(1) Similar content tags

The system reads the title, text keywords, participating topics and other information to roughly determine the content tags. It has been observed that in the algorithm for determining note similarity, the same topic and similar keywords are given high weights. (If it is a video note, click on the homepage of the APP and scroll down to see similar videos. If it is a picture note, share the note to the WeChat file transfer assistant, then click to enter the share, and there will be related content recommendations under the note. If you are interested, you can observe it yourself.)

When there are enough notes with similar text to the note under this topic, basically all the notes under this topic will be recommended. If there are not enough similar notes under the same topic, notes with similar keywords under other topics will be recommended.

(2) Image similarity

Similar images should also be among the influencing factors. The specific crawling logic may be that when there is no note content with similar keywords in the recent period, content with relatively similar images will be selected for recommendation. Image similarity is only for graphic notes.

Therefore, content creators need to select keywords well, understand the characteristics of similar notes under the keyword, and strive to obtain better interaction data when the system recommends it for the first time, so as to enter the next traffic pool faster. In addition, you can constantly learn about the target user profile from these similar notes and bloggers, and adjust the account's content positioning and content production strategy more quickly.

If you still have confusion about Xiaohongshu promotion and need answers, you can read "Answers to 31 Difficult Questions about Xiaohongshu Monetization and Operations!" for details. It is recommended to collect the article "Hope it can help you!

Author: Zhang Ling

Source: Zhang Ling