Job Description

About the Team: CCR & VOC

The Consumer Complaint Rate (CCR) and Voice of Customer (VOC) team is at the heart of understanding and improving the user experience on TikTok's e-commerce platform.

Consumer Complaint Rate (CCR): This metric measures negative consumer feedback by using Natural Language Processing (NLP) to interpret texts from various channels, including product reviews, service tickets, and video comments. It provides critical insights into governance and experience issues across products, customer service, logistics, and content.

Voice of Customer (VOC): This initiative analyzes the original voice of consumers to proactively identify issues affecting user experience. By monitoring product reviews, video reports, and complaint tickets, VOC integrates alarm information, transfers details to the appropriate teams for resolution, and tracks subsequent trends.

Our vision is to build an intelligent, efficient, and AI-driven future for e-commerce feedback analysis. We are replacing traditional manual processes with an advanced AI-powered system that leverages Large Language Models (LLMs) and NLP. This system will provide instant, seamless support by understanding user feedback, identifying problems, and delivering actionable analysis to enhance the overall shopping experience without human intervention.

Join us to be at the forefront of this revolution. You will help build an AI system that not only understands and adapts to user feedback but also provides intelligent solutions, ultimately reducing costs and increasing efficiency for our platform.

Responsibilities:

- Develop LLM-based CCR/VOC Models: Design and implement LLM-based CCR/VOC models, to cover user feedback from multiple channels, including reviews, after-sales service, customer service conversations, and tNPS, etc. Based on a 100+ intent classification system, identifies the user's feedback intent accurately.

- Apply Advanced LLM Techniques: Utilize state-of-the-art LLM post-training techniques, such as instruction tuning, Reinforcement Learning from Human Feedback (RLHF), and continuous learning, to optimize the system with minimal labeled data.

- Build Robust Training Datasets: Identify challenging user feedback scenarios—including logistics inquiries, merchant issues, and platform-related problems—and construct specialized datasets to enhance AI model training.

- Enable Multilingual and Multicultural Support: Build AI models capable of recognizing user feedback across diverse languages and cultural backgrounds, ensuring accurate semantic understanding for a global audience.

- Optimize Model Efficiency and Deployment: Research and apply model compression, quantization, and efficient inference techniques to ensure the AI assistant operates at scale with low latency and high reliability.

- CCR/VOC RCA Agents： Focusing on pain points in CCR, such as the need for rapid implementation of new labels and inefficient CCR problem handling processes in business side, this project leverages the agent's autonomous interaction and intelligent decision-making capabilities to optimize existing models and processes. Through scenario-based training and iteration, the agent becomes a highly efficient assistant for CCR operations, solving the problems of slow response and low conversion rates in traditional models, and improving service quality and customer experience.