In China’s rapidly evolving short-video landscape, understanding how each platform recommends and reviews content is key to mastering visibility and engagement.
While Douyin (Tiktok), Xiaohongshu, and WeChat Video may all share the goal of connecting creators with audiences, their underlying algorithms and traffic systems operate very differently.
From Douyin’s deep-learning behavior prediction, to Xiaohongshu’s CES-driven community logic, and Video Account’s social-powered distribution, each platform represents a distinct ecosystem shaped by technology, user intent, and social context.
This article breaks down how these three major players analyze text, images, and interactions to drive content exposure revealing the mechanisms behind what makes a post go viral, stay relevant, or quietly fade into the algorithmic void.
Douyin uses a deep learning model + decentralized recommendation mechanism to predict user behavior through neural networks.
Douyin’s content distribution process is mainly divided into three stages:
1. Video Review Stage
Before going live, every video passes AI screening for voice, subtitles, and visuals. The system filters out violations like copyright issues, vulgar or misleading content. Once cleared, it enters the cold-start phase – shown to 200 – 500 initial viewers.
2. Algorithm recommendation stage
Through multimodal feature recognition technology, in-depth analysis of video content
The platform uses a “scoring mechanism” to judge the subsequent recommendations of the video, and comprehensively calculates user behavior indicators such as completion rate, like rate, comment rate, and forwarding rate. Douyin’s recommendation algorithm no longer relies on labeling content and users. Instead, it directly predicts user behavior through neural networks and calculates the total value users gain from watching content.
3. Traffic Allocation Mechanism
Douyin’s algorithm avoids repetitive AIGC (AI-generated Content) content by mixing diverse videos in your feed.
It also uses multi-interest recall – analyzing watch time, search history, and comments to uncover your hidden interests, not just the obvious ones.
Xiaohongshu’s recommendation algorithm is based on the CES score (community engagement score)
Adopting a two-way matching mechanism of “content tags + user tags“
The CES scoring formula is: CES = number of likes (1 point) + number of favorites (1 point) + number of comments (4 points) + number of reposts (4 points) + number of followers (8 points)
Content recognition process:
On Xiaohongshu, engagement is everything.
Likes, comments, and saves determine whether your post reaches 10K, 100K, or even 1M users.
Because the platform has a long recall cycle, even posts from months ago can resurface – so focus on quality visuals, useful content, and strong keywords to keep your post discoverable through search.
The recommendation algorithm of WeChat Channels is fundamentally different from that of Douyin, Kuaishou and other platforms. Its core logic is “private domain traffic leverages public domain traffic”. The content weight is the lowest among the three major platforms, accounting for about 50%, while the social relationship chain weight is relatively high.
Recommendation mechanism:
Content review process:
WeChat Video Rules & Traffic
WeChat Video enforces stricter checks on live content – banning recorded streams, fake info, and low interaction.
The ranking weight of content scores is: completion rate > number of likes > number of comments > number of clicks on extended links > number of reposts > number of favorites.
Douyin: Sub-word Segmentation and Deep Semantic Understanding
Douyin’s text NLP processing uses subword segmentation technology, mainly based on the following methods:
Xiaohongshu’s Keyword Strategy
Xiaohongshu focuses on keyword placement and search optimization. It uses a dictionary-based segmentation algorithm and follows the “70% long-tail keyword rule” to boost discoverability. Tags combine manual inputs with NLP extraction, adjusting recommendations in real time based on user behavior.
WeChat Video: Smart Tagging System
For video content detection, AI will combine text (subtitles, titles, video introductions) with NLP word splitting. For example, “Shanghai Oriental Pearl Tower” will be split into two sets of words: “Shanghai” and “Oriental Pearl Tower”. Images (covers) and features of the cover, such as faces and clothing, are all detection targets. Sound effects (background music), video images (frame detection) and other multimodal features are used for compliance review.
Generative AI model applications: such as S-YOLO V5 and Vision Transformer models for video content description generation, combined with attention mechanisms to enhance keyframe recognition and improve text generation quality.
Thank you for exploring our content. We aim to provide valuable insights into the Chinese market to aid your decision-making and support your business at every step.
Our blog is your resource for Chinese marketing tips and Chinese market guidance. Contact us with questions or for more service details below.
Website: www.mmgthailand.com
Tel: 06-3167-8206
Email: info@mmgthailand.com
FB: MMG Thailand
IG: mmg_thailand
About MMG Thailand:
MMG Thailand is the first and only Chinese-owned Chinese marketing company in Thailand that aims to connect Sino-Thai cultures and power partners’ success.
We:
We’re committed to providing deep insights into the rapidly changing Chinese market and leveraging our rich experience to facilitate your growth in this dynamic environment.
Chinese outbound travel discovery is increasingly driven by interests, travel scenarios, and intent not simply by destination names. At the center of this shift is Xiaohongshu (XHS), which has become an increasingly influential.
You may have heard this claim when exploring Xiaohongshu (XHS) marketing. It sounds convincing, but it’s not quite how Xiaohongshu works.
China’s luxury market is entering a new phase. As affluent Chinese consumers shift their spending from luxury goods toward travel, wellness, and meaningful experiences, jewelry brands are facing a tougher challenge: retaining valuable Chinese customers in a slower market.
Contact us
Due to high demand of our services, we offer only a 40-minute free consultancy session
Soi sukhumvit 39 Khlong Toei Nuea, Bangkok 10110
211 Soi Pridi Banomyong 11, Sukhumvit 71 Rd., Khwaeng Phra Khanong Nuea, Watthana, Bangkok 10110
Influencer (KOL/KOC) Marketing
Xiaohongshu (Redbook) Marketing