Machine Learning Engineer (Content Ecology & Creator) - E-commerce Governance
TikTok
Posted 12 days ago
About this role
About the Team
Building a Prosperous, Trusted, and Fair Global E-Commerce Ecosystem
We are the Governance & Experience Algorithm Team, the AI guardians ensuring the long-term health of TikTok Shop’s global platform. As our international business expands, our mission goes beyond traditional risk control. We are dedicated to constructing a prosperous, trusted content ecosystem and maintaining a fair, healthy environment for creators.
We leverage LLM agents, RAG, GNN, and Sequence Modeling to solve complex governance challenges. We don't just block bad actors; we shape the rules of the game to ensure that creativity is rewarded, fairness is upheld, and the ecosystem thrives.
Our Core Mission:
- Trust & Quality: Ensuring users trust what they see, establishing a standard where "Good Content = Good Business."
- Creator Governance: Managing the full lifecycle of creators by identifying malicious intent (e.g., piracy, content mills) while protecting high-potential authentic creators.
- Ecosystem Fairness: using AI to ensure fair traffic distribution and prevent monopolies by bad actors, fostering a diverse and sustainable creator community.
What You’ll Do
1. Creator Governance & Quality Modeling
- Signal-Driven Creator Profiling: aggregated underlying multi-modal signals (e.g., static frames, low-aesthetic detection, piracy fingerprints) to build comprehensive Creator Quality Scores.
- Combat Low-Quality & Malicious Intent: Develop sequence-based models to detect and penalize creators engaging in "low-effort selling," "re-recording/piracy," and "matrix account spamming," effectively purging the ecosystem of noise.
- LLM & RAG Intelligent Governance: Build LLM + RAG systems that dynamic interpret complex governance policies. Develop agents that not only flag risky creators but provide explainable reasoning to guide creator education and improvement.
2. Graph Intelligence & Syndicate Detection
- Heterogeneous Graph Mining: Construct large-scale Heterogeneous Graphs (Creator-Product-Video-User) to uncover hidden relationships and organized bad actors (e.g., fake engagement rings, black-market account trading, sybil attacks).
- Cross-Domain Risk Propagation: Utilize graph algorithms to track how risk propagates across different scenarios (Content vs. Shelf) and markets, predicting where bad actors will migrate next.
3. Ecosystem Strategy, Fairness & Optimization
- Multi-Objective Optimization (MMoE/PLE): Develop advanced multi-task learning models to balance conflicting objectives—maximizing Ecosystem Prosperity and GMV while minimizing Governance Risk and User Complaints.
- Fairness Algorithms: Design traffic regulation strategies that prevent the "rich get richer" effect for low-quality diverse content, ensuring fair exposure for high-quality, original creators.
Building a Prosperous, Trusted, and Fair Global E-Commerce Ecosystem
We are the Governance & Experience Algorithm Team, the AI guardians ensuring the long-term health of TikTok Shop’s global platform. As our international business expands, our mission goes beyond traditional risk control. We are dedicated to constructing a prosperous, trusted content ecosystem and maintaining a fair, healthy environment for creators.
We leverage LLM agents, RAG, GNN, and Sequence Modeling to solve complex governance challenges. We don't just block bad actors; we shape the rules of the game to ensure that creativity is rewarded, fairness is upheld, and the ecosystem thrives.
Our Core Mission:
- Trust & Quality: Ensuring users trust what they see, establishing a standard where "Good Content = Good Business."
- Creator Governance: Managing the full lifecycle of creators by identifying malicious intent (e.g., piracy, content mills) while protecting high-potential authentic creators.
- Ecosystem Fairness: using AI to ensure fair traffic distribution and prevent monopolies by bad actors, fostering a diverse and sustainable creator community.
What You’ll Do
1. Creator Governance & Quality Modeling
- Signal-Driven Creator Profiling: aggregated underlying multi-modal signals (e.g., static frames, low-aesthetic detection, piracy fingerprints) to build comprehensive Creator Quality Scores.
- Combat Low-Quality & Malicious Intent: Develop sequence-based models to detect and penalize creators engaging in "low-effort selling," "re-recording/piracy," and "matrix account spamming," effectively purging the ecosystem of noise.
- LLM & RAG Intelligent Governance: Build LLM + RAG systems that dynamic interpret complex governance policies. Develop agents that not only flag risky creators but provide explainable reasoning to guide creator education and improvement.
2. Graph Intelligence & Syndicate Detection
- Heterogeneous Graph Mining: Construct large-scale Heterogeneous Graphs (Creator-Product-Video-User) to uncover hidden relationships and organized bad actors (e.g., fake engagement rings, black-market account trading, sybil attacks).
- Cross-Domain Risk Propagation: Utilize graph algorithms to track how risk propagates across different scenarios (Content vs. Shelf) and markets, predicting where bad actors will migrate next.
3. Ecosystem Strategy, Fairness & Optimization
- Multi-Objective Optimization (MMoE/PLE): Develop advanced multi-task learning models to balance conflicting objectives—maximizing Ecosystem Prosperity and GMV while minimizing Governance Risk and User Complaints.
- Fairness Algorithms: Design traffic regulation strategies that prevent the "rich get richer" effect for low-quality diverse content, ensuring fair exposure for high-quality, original creators.