Traffic password for Xiaohongshu's popular notes

I recently discovered an interesting phenomenon - people often report on Xiaohongshu that no one reads the carefully planned content, but sometimes a random post can get a lot of traffic.

Why is this?

There are many opinions on the Internet about the traffic distribution of Xiaohongshu:

There is a fixed note scoring formula inside Xiaohongshu;

Xiaohongshu requires account maintenance. If the account weight is not high, the traffic of the notes will be very small.

The content of the note needs to repeat the keywords multiple times to increase the chance of being searched...

Do these statements make sense? What exactly is Xiaohongshu’s traffic distribution mechanism?

With these questions in mind, the operation agency conducted research on Xiaohongshu's traffic mechanism. At the same time, it also talked with @勺子, the content manager of Anixingkong, an MCN agency that has won the TOP 1 on the Xiaohongshu Sports List many times, and summarized some key points.

01 What is the traffic mechanism of Xiaohongshu?

After a note on Xiaohongshu is published, it can be viewed through four pages, namely the follow page on the homepage, the discovery page, the local page, and the search page to which the search button can jump.

Among them, the discovery page and search page are the two main traffic entrances. Taking the notes of @Luuuuke, a user on Jike as an example, the traffic from homepage recommendations and searches accounted for 40% each.

Image source: @Luuuuke

According to the observation of the operator, these two pages have different traffic distribution logic.

1) Recommendation logic of the discovery page

On the discovery page, notes are presented in the form of an information flow. Every time the user pulls down to refresh, a new batch of notes will appear. These notes are mainly displayed through recommendation algorithms.

The operation agency referred to two publicly shared articles by @赵晓萌, the then algorithm architect of Xiaohongshu in 2017, and @郭一, the then head of Xiaohongshu's real-time recommendation team in 2019, as well as a paper released by Xiaohongshu's algorithm team in 2021, and analyzed the recommendation logic of the discovery page.

The note recommendation process can be roughly divided into three steps:

First, pre-score based on past user data before recommendation, then make recommendations based on user preferences, and then put the notes into an increasing traffic pool based on user interaction feedback.

① Estimate the effect before making a recommendation

According to @郭一's description, notes will not be recommended as soon as they are published, but will first go through a round of "scoring" by the recommendation system before being recommended.

In the first step, the system will find similar notes from the past note database.

In the second step, the system will predict the probability of receiving user interaction after the note is published based on the interactive data such as clicks, likes, and collections obtained by similar notes, and thus "score" the note.

In other words, by analyzing similar notes, it is possible to predict whether this note will be liked by users after it is sent.

In the algorithm for estimating ratings, the probability of a user clicking has the greatest impact on the rating. The probabilities of other interactive behaviors are assigned a weight, and the rating of the note is ultimately calculated.

For example, assuming that we only consider clicks, likes and collections, and the weights of likes and collections are both 0.5, the estimated click probability of a note is 50%, and the like and collection probabilities are both 10%, then the final score of this note is 0.05 points.

The algorithm of the recommendation system is constantly updated through feedback from existing users, so the scoring formula is also constantly changing.

Finally, the system will adjust the position of the notes according to the score.

The operation agency speculates that notes with high scores will be positioned higher in the information flow and receive more exposure, while notes with low scores will be positioned lower and receive less exposure.

②Distribute content according to user preferences

When a note is put into online recommendation, Xiaohongshu will match the content tags of the note with the content tags preferred by the user.

According to what @赵晓萌 shared in 2017, when a publisher posts a note, Xiaohongshu's algorithm will recognize the text and images in the note and label them as content.

One special point is that the algorithm can also classify emotion-related words, identify which notes with emotion tags users like to browse, and match content with similar emotions.

For example, the operation agency found that after frequently searching and liking joke notes, Xiaohongshu will start to recommend more notes that also express "happy and joyful" emotions.

At the same time, based on dimensions such as topics of interest, favorite albums, and the degree of user interaction with notes, Xiaohongshu can also infer the content tags preferred by users.

User and note labels in 2017 (current labels may have changed)

Through matching, content can be recommended to the homepage of users who may be interested in it.

Based on the behavior of user accounts, Xiaohongshu's recommendation and search data are also interconnected:

After the user searches for the keyword "xx", the system will think that the user needs to know "xx" and will push relevant content on the discovery page. For example, if you search for "Lantern Festival", the homepage will recommend "Tangyuansu".

However, if the user does not continue to browse relevant content, the system's recommendations will stop.

At the same time, Xiaohongshu mainly makes recommendations based on user preferences, but it will also try to ensure the diversity of the results and recommend some notes that users do not prefer but have higher ratings.

③Interaction determines whether to recommend

After Xiaohongshu pushes notes to an initial batch of users based on user preferences, it will decide whether to recommend it to more people based on the interaction data of this group of users, such as likes, favorites, and comments.

The better the interaction data, the more recommendations the notes will receive and enter an increasingly expanding traffic pool.

After a period of verification, the operation agency found that the recommended notes on the homepage are generally published within the past two months. However, the amount of interaction with notes varies depending on the time.

For notes that were published a short time ago, the number of likes and interactions may vary. For example, in the screenshot, there are two notes that were published on the same day, but one has 49,000 likes and the other only has 28.

Posts that have been published for a longer time generally have more likes and interactions. For example, three of the notes in the screenshot were posted a month ago, but they all received over a thousand likes.

From this, we can infer that in the short period after a note is published, regardless of the amount of interaction, it is likely to be recommended. But if there is no continued user interaction, the system will stop recommending the note.

Therefore, notes that were published a long time ago and have a small number of interactions will basically not be presented in the homepage recommendations.

Only when users continue to interact will the notes enter the increasing traffic pool and be continuously recommended over a long period of time.

2) Sorting logic of search pages

In addition to the discovery page, the search page is also a large traffic entrance.

According to data released at the Xiaohongshu Business Conference not long ago, 30% of Xiaohongshu users will start searching directly after entering the APP.

On the discovery page, Xiaohongshu mainly uses recommendation logic, while on the search page, Xiaohongshu allocates traffic according to the sorting logic. The higher the ranking of the notes, the greater the exposure they will receive.

But this order is not fixed, and the order of notes is constantly changing with the real-time calculation of the algorithm.

@勺子 told the operator that on the search page, there are three main factors that affect the ranking results: the degree of keyword matching, the amount of interaction in a short period of time, and the viewer's account behavior.

①Keyword matching degree

In Xiaohongshu search, the degree of match between the notes and the keywords searched by the user will greatly affect the ranking position of the notes in the search results page.

When a user searches for a keyword, the system prioritizes notes that closely match the keyword.

For example, when searching for the keyword "Gongqing Park Cherry Blossoms", the note ranked first on the search page does not have as much interaction as other notes, but the content of the note matches "Gongqing Park Cherry Blossoms", so it will be displayed first.

Keywords can be placed in the title, content, and topic of a note.

So how do you arrange the keywords so that your notes are ranked higher?

@勺子 said that mentioning recently popular keywords, or words that users in related fields will pay attention to in the title and content will indeed increase the probability of the notes being searched. It doesn’t matter if you mention the keyword multiple times.

The so-called "words that users will pay attention to in related track fields" refer to the "subdivided keywords" under this topic field. There are two reasons why we need to layout "segmented keywords".

First of all, if the scope of a keyword is too broad, there will be more notes under that keyword, the competition will be too great, and it will be more difficult to get traffic.

After testing, the operation agency found that the more "specific" the search keywords are, the smaller the changes in the ranking of the note results will be.

For example, the operation agency searched for a more "specific" keyword "blue and white British shorthair price" one month later, and the top two notes on the search page from the previous month still remained in the front position.

When searching for the broader keyword "beauty", it was found that the top three notes had changed in just 4 minutes.

Secondly, when users search, they tend to search for a more precise keyword. For example, when users want to know about "blue and white British shorthair", they will search for "the price of blue and white British shorthair", "how to raise a blue and white British shorthair", etc.

Therefore, the more precise the keywords in the notes are, the easier it is for users to search for them, the longer they will remain on the results page, and the more traffic they will receive.

②The amount of interaction in a short period of time

Another factor that affects the ranking of notes in search results pages is the amount of interaction the note receives within a short period of time after being published.

@勺子 told the operation agency that if a note can gain a lot of interactions (likes, favorites, comments, etc.) within a short period of time after being published, it means that it has the potential to become a hit, and it will be ranked higher in the search results page.

For example, after searching for "Shanghai Cherry Blossoms", the operation agency found that there were several notes with 200-400 likes, which were ranked above notes with around 1,000 likes, and the content was all related to "Shanghai Cherry Blossoms".

Screenshot taken on March 14

The operating company speculates that this may be related to the amount of interactions the note received within the time of publication.

The first note in the picture below was published around March 12, and received more than 400 likes in just two or three days, and has the potential to become a "hit"; the next two notes with about 1,000 likes were published as early as the first half of 2021, about a year apart from now.

③Viewer’s account behavior

An operations operator @赵子辰Vic once mentioned that Xiaohongshu’s keyword search results are “different for different people” and present different results to different users.

The sorting algorithm will infer the user's understanding of the topic based on the user's past behavior, thereby matching different search content.

When the operator uses different accounts to search for the same keyword at the same time, the pages presented are different.

For example, when searching for "food" at the same time, Xiaohongshu shows more Shanghai restaurant exploration notes for an account (left side of the picture), and the content displayed on the right side of the picture includes food preparation and sharing, and the results are more diverse.

The operator speculates that the user of the account on the left may have recently searched and browsed Shanghai restaurant exploration notes, so the system will infer that the user searched for "food" because he wants to know more about which restaurants are available in Shanghai.

02How to improve operational capabilities under the traffic mechanism of Xiaohongshu?

So what specific impact will Xiaohongshu’s traffic distribution mechanism have on creators? How to adjust operational ideas based on the traffic distribution mechanism?

1) Note content needs to provide value to users

The most important impact of Xiaohongshu's traffic mechanism on creators is that the content must be able to provide value to users.

Only when users feel the value of the notes will they like and collect them. Therefore, the notes can enter the increasing traffic pool and be ranked higher in the search results page.

@勺子 said that in the Xiaohongshu community, the more popular content generally provides two types of value: one is practical value, and the other is emotional value.

The so-called practical value means that the notes can provide users with practical knowledge.

For example, this note, which has received more than 20,000 likes and collections, revolves around a problem that some people often encounter in life: "What should I do after losing my phone?" and provides practical knowledge.

@勺子 also believes that in addition to sharing practical knowledge, more popular notes will also give users "emotional value."

The so-called notes with "emotional value" refer to some notes (such as chatting) that may not be practical, but can convey an emotion to the user (such as happiness, sadness, etc.).

For example, this note about a couple's quarrel conveys a sentiment of "sharing funny daily life", which can arouse the attention of users who have the same experience and stimulate interest in discussion.

This note received over 40,000 likes and over 3,000 comments, with many users responding in the comment section saying “this is so funny”.

At the same time, the operator believes that Xiaohongshu’s traffic distribution mechanism also ensures that more content has the potential to be seen. As long as the content of the note is valuable in the community, even if it is not popular now, it may become popular later.

@勺子 also said that if the content keywords become hot words, the accumulated high-quality notes can regain traffic through search.

For example, during this year's Winter Olympics, after Bing Dwen Dwen became popular, a note about "Bing Dwen Dwen" published in 2019 also regained traffic.

Focusing on user value also requires, on the other hand, avoiding producing content that is inconsistent with community values, such as showing off wealth, skirting the rules, and maliciously provoking conflict. There are also corresponding reminders on the note publishing interface of Xiaohongshu.

@勺子 said that if there is any illegal content, the note will not be recommended. As long as the notes comply with the regulations, they will be included in the system.

2) Focus on niche areas and create original vertical accounts

It is mentioned in many Xiaohongshu "operation strategies" that the publisher's account weight will affect Xiaohongshu's traffic distribution, but the official has denied this.

In the notes posted by @土豆小助手, it was emphasized that Xiaohongshu has no weight in the account dimension.

@勺子 also explained that when creating an account on Xiaohongshu, there is no need to "maintain the account" (increasing the "account weight" through normal user browsing, liking, etc.), but the originality of the account and the verticality of the content need to be guaranteed.

In addition to avoiding plagiarism, ensuring originality also requires avoiding content homogeneity. Operators need to explore users’ pain points, produce content from different perspectives, and deliver it to targeted user groups.

For example, the "shoulder opening movement" that yoga bloggers are doing, if you simply teach this movement, it may be very similar to other people's content. But if you produce content on how to "do shoulder-opening exercises while lying down" and approach it from a different perspective, the effect may be better.

The verticality of the account content will affect the distribution of the content.

@勺子 mentioned that both creators and users have "labels", and the content verticality of an account is part of the creator's "label". The creator's content will first be pushed to users who match his or her tags.

For example, if a creator’s tag is “fitness”, then his/her notes will be pushed first to users whose “tags” are “fitness, sports”. After seeing the post, the user is more likely to like or collect it.

This also means that if the account content is not vertical, the group of people matched with the notes will be more confusing, resulting in fewer effective interactions and a smaller possibility of being continuously recommended.

Therefore, when positioning an account, you need to think about which groups of people the content is targeted at and how to provide more value to these users through content production.

At the same time, @勺子 also shared that notes that follow hot topics on Xiaohongshu will indeed gain some topic traffic, but whether to follow this topic still depends on whether the topic and the account match.

If you affect the verticality of your account in order to chase hot topics, it will be counterproductive.

3) Pay attention to deep interaction data

Since the interactive scoring mechanism is very important in Xiaohongshu's traffic distribution, creators need to think about how to guide interaction based on the content of the notes.

In 2017, @赵晓萌 shared that in the community, Xiaohongshu's content algorithm measures the depth of interaction in the community, citing basic interaction data such as clicks, likes, and favorites as examples.

However, the operation company believes that with the development of Xiaohongshu’s algorithm, this “deep interaction” should also be understood at a deeper level. In other words, deep interaction not only includes basic likes, favorites, and comments, but also includes further interactive actions.

For example, "attention brought by notes" belongs to deep interaction.

@勺子 believes that in the scoring mechanism for note recommendations, attention is a very important point, and this number of attention does not refer to the number of fans the blogger has, but the number of fans brought by the notes.

At the same time, the operating agency speculates that the "floor within a building" formed by multiple replies to comments in the comment section of Xiaohongshu notes is also a kind of deep interaction.

It has been observed that the notes recommended on the homepage of Xiaohongshu and the notes ranked high in the search interface generally have more "in-depth" discussions among users in the comment section, and some hot comments can trigger hundreds of replies.

In other words, in addition to working hard on the value of notes and the verticality of accounts, it is also possible to obtain more traffic by attracting user attention in notes, maintaining the activity in the comment area, and improving the "deep interaction data" of notes.

03 Conclusion

Xiaohongshu mainly has two traffic entrances: the discovery page and the search page. The discovery page follows the recommendation logic, while the search page follows the sorting logic.

The content presented on each page will vary due to the different historical behaviors of the reader's account.

But the similarity is that operators need to explore the pain points of target users, produce notes with practical or emotional value, and focus on data on deep interactions.

At this level, actually doing it is the really difficult part, and understanding Xiaohongshu's traffic distribution mechanism is just the beginning.

PS Since we cannot know the detailed algorithm logic, the article is for reference only. Everyone is welcome to think about it while reading and discuss it with us.

Not only Xiaohongshu, the traffic distribution mechanism of most platform-based apps is closely related to the commercialization of the platform.

Author: Operation Research Society

Source: Operation Research Society