Live Streaming Recommendation Algorithm
1.Sources of Traffic:
Types of Traffic Sources:
A:System Recommendations: Traffic driven by algorithmic suggestions from the platform.
B:Fan Relationships: Traffic coming from existing followers and their interactions.
By analyzing the trends in traffic from different sources, you can quickly identify the main issues if something goes wrong.
Conversely, if traffic from short video recommendations decreases, it might be because fewer short videos are being posted or the posting timing is off. Understanding these traffic sources and their analysis is a fundamental skill for managing a streamer’s operations.
The traffic sources vary greatly depending on the streamer’s level.
To diagnose unusual traffic patterns, compare current data with historical trends and also benchmark against similar streamers (with comparable follower counts, revenue levels, and content niches).
Case Study Analysis:
2.How the System Allocates Traffic
A. Initial Traffic:
When a streamer begins a broadcast，the system sends an initial burst of traffic known as cold start traffic.
The system uses data from this initial traffic to assess:
The system uses data from this initial traffic to assess:
The suitability of the content for different types of users.
The performance during this cold start phase impacts the full-cycle traffic allocation.
Therefore, streamers need to be well-prepared and deliver engaging content right from the start. Failing to do so—like airing meaningless content or a blank screen—can negatively affect future traffic recommendations.
B. Full-Cycle Traffic:
After the cold start phase, the system shifts to providing full-cycle traffic. TikTok’s goal is to continuously meet user demand for live content by finding the best live streams from the content pool. In short, it aims to discover the streams that users enjoy the most.
Step1: The system calculates an overall satisfaction index based on various performance metrics:
Positive Indicators: Click-through rate (CTR), effective viewing rate (CTR for 10 seconds), average watch time, likes, comments, follow rate, shares, gift rate, etc.
Negative Indicators: Dislikes, report counts, dislike rate, report rate, negative feedback from in-feed surveys, etc.
Step2: Based on this satisfaction index, the platform recommends content to users. However, because the recommendation system can struggle to differentiate between good and bad content, both automated and manual reviews are conducted to ensure content quality before it is distributed to viewers.
EN 
EN_GB 
AR 
DE 
ES 
FR 
ZH 
JA 
ID 
KO 
TH 
VI 
