Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Chinese Morph Resolution in E-commerce Live Streaming Scenarios

Jiahao Zhu, Jipeng Qiang, Ran Bai, Chenyu Liu, Xiaoye Ouyang

TL;DR

The paper tackles morph-based deception in Chinese health and medical e-commerce live streaming by introducing LiveAMR, a pronunciation-focused morph resolution task. It formulates LiveAMR as a text-to-text generation problem using ASR transcripts and a Mengzi-T5 backbone, augmented with LLM-generated data to improve robustness across ASR variations. A large-scale LiveAMR dataset (86,790 samples, 2,688 morphs; 431 original words) is constructed via multi-stage annotation and filtering, and a regulatory-focused CLiveSVD dataset is used to assess downstream impact. Experimental results show that morph resolution significantly boosts detection accuracy and regulatory effectiveness, with data augmentation yielding substantial gains and broader generalization, highlighting practical value for consumer protection and platform standardization. The work also discusses limitations and future directions for expanding topic coverage and validating across more models.

Abstract

E-commerce live streaming in China, particularly on platforms like Douyin, has become a major sales channel, but hosts often use morphs to evade scrutiny and engage in false advertising. This study introduces the Live Auditory Morph Resolution (LiveAMR) task to detect such violations. Unlike previous morph research focused on text-based evasion in social media and underground industries, LiveAMR targets pronunciation-based evasion in health and medical live streams. We constructed the first LiveAMR dataset with 86,790 samples and developed a method to transform the task into a text-to-text generation problem. By leveraging large language models (LLMs) to generate additional training data, we improved performance and demonstrated that morph resolution significantly enhances live streaming regulation.

Chinese Morph Resolution in E-commerce Live Streaming Scenarios

TL;DR

The paper tackles morph-based deception in Chinese health and medical e-commerce live streaming by introducing LiveAMR, a pronunciation-focused morph resolution task. It formulates LiveAMR as a text-to-text generation problem using ASR transcripts and a Mengzi-T5 backbone, augmented with LLM-generated data to improve robustness across ASR variations. A large-scale LiveAMR dataset (86,790 samples, 2,688 morphs; 431 original words) is constructed via multi-stage annotation and filtering, and a regulatory-focused CLiveSVD dataset is used to assess downstream impact. Experimental results show that morph resolution significantly boosts detection accuracy and regulatory effectiveness, with data augmentation yielding substantial gains and broader generalization, highlighting practical value for consumer protection and platform standardization. The work also discusses limitations and future directions for expanding topic coverage and validating across more models.

Abstract

E-commerce live streaming in China, particularly on platforms like Douyin, has become a major sales channel, but hosts often use morphs to evade scrutiny and engage in false advertising. This study introduces the Live Auditory Morph Resolution (LiveAMR) task to detect such violations. Unlike previous morph research focused on text-based evasion in social media and underground industries, LiveAMR targets pronunciation-based evasion in health and medical live streams. We constructed the first LiveAMR dataset with 86,790 samples and developed a method to transform the task into a text-to-text generation problem. By leveraging large language models (LLMs) to generate additional training data, we improved performance and demonstrated that morph resolution significantly enhances live streaming regulation.
Paper Structure (15 sections, 3 equations, 4 figures, 6 tables)

This paper contains 15 sections, 3 equations, 4 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Task Definition
  4. Dataset Construct
  5. Methods
  6. Experiment
  7. Experimental Setup
  8. Implementation Details
  9. Experimental Results
  10. Usefulness of Morph Resolution
  11. Ablation Study
  12. Conclusion
  13. The annotation Website
  14. Prompt templates in this paper
  15. More Examples

Figures (4)

  • Figure 1: Example of morph used in the live streaming scenarios
  • Figure 2: Performance with different number of training samples.
  • Figure 3: Screenshot of an annotation example on the annotation Website. The red text indicates added comments.
  • Figure 4: The prompting template of generating sentences. Generate context-appropriate sentences that contain the specified vocabulary and meet the required quantity.