Introduction

This report focuses on analyzing the recommendation system of the currently popular Chinese social app, Rednote(Xiaohongshu). Rednote is a social commerce platform containing lifestyle content, e-commerce, enteraintment, and so on. I personally use a lot of this app, and what it is different with another Chinese social app, Douyin(Chinese Tiktok), is that it’s more like a community platform in which users can share posts with a mix of videos, images, or texts purely, while Douyin is a pure entertainment platform of short videos.

The goal of this report is to: 1. Apply a scenario design to better understand the users and the company’s goals. 2. Reverse engineering to analyze how Rednote’s recommendation system may work. 3. Propose recommendations for improvement.

Scenario Design

Three questions need to be answered: 1. Who are the target users? 2. What are their goals? 3. How does the recommendation system help to achieve the goals?

These three questions would be asked and answered in two perspectives: the end users and the company.

User Perspective

  1. Who are the target users?

    The primarily target users are young adults, aged from 18 to 35. The users are usually urban, receiving mid to high education, and having strong interest in beauty, lifestyle, travel and fashion, among which female users might also seek products recommendations.

  2. What are their goals?

    Users want to explore more lifestyle ideas, travel plannings, social connections and products recommendations.

  3. How does the recommendation system help to achieve the goals?

    The system promote relevant, high-quality contents to the users based on their activities, such as likes, comment,or dislikes.

Company Perspective

  1. Who are the target users?

    The company’s customers include end users, creators and advertisers.

  2. What are their goals?

    The company’s goals are to increase user engagement, retain users and eventually monetize through advertisments, partnerships and e-commerce.

  3. How does the recommendation system help to achieve the goals?

    The system helps to achieve the goals by: promoting relevant and emotionally content to increase engagement; promoting high-quality creators to maintain the community; connecting contents to products.

Reverse Engineering

  1. Engineers may use machine learning models like CLIP to classify notes into different topics, such as emotion, travel, beauty, etc.

  2. Build interest graphs from likes, saves, search items, dislikes, reports; build social graphs from follows, comments, chats.

  3. Stages:

    1. retrieve posts from pools.
    2. filter by relevance.
    3. ranking by likes, saves, etc.

  4. Two types of feedbacks:

    1. positive: likes, saves, comments, shares.
    2. negative: dislikes, quick switchs, skips, fast scrolls, reports.

Recommendations

I would recommend adding a user adjustable controller which allows to control the amount of promoting contents, such as more cars, less emotion, no marriage. Also, they can also connect to adjacent interests, for example, if the user likes piano, contents like how to maintain a piano or how a piano is built could be feeded.

Also, in my personal experience, I prefer longer, higher-quality contents intsead of constant scrolling-down and switching.

Conclusion

Rednote’s recommendation system is sophisticated which balances the needs of company, end users, creators and partners. The community-like platform attracts users who want to share their lifestyles or products. I believe future maintenance in algorithms, user control harmonious community will improve both the users and the company’s satisfaction.

References

Zhang, Y., & Zhao, W. (2023). Hybrid Recommendation Strategies in Chinese Social Platforms: Case Study of Xiaohongshu. Journal of Information Systems Research.

Public UX observations and user reports from Xiaohongshu (App Version 8.x, 2025).