Related Experiment Video
Updated: Sep 13, 2025
09:27
Using Eye Movements Recorded in the Visual World Paradigm to Explore the Online Processing of Spoken Language
Published on: October 13, 2018
10.2K
LLM-Enhanced Chinese Morph Resolution in E-Commerce Live Streaming Scenarios
Xiaoye Ouyang1, Liu Yuan1, Xiaocheng Hu1
1China Academy of Electronic and Information Technology, Beijing 100041, China.
Entropy (Basel, Switzerland)
|July 29, 2025
Summary
This study introduces a new method for resolving morphed speech in e-commerce live streams, improving accuracy in detecting misleading claims. The approach uses large language models to enhance speech transcription for better consumer protection.
Area of Science:
- Natural Language Processing
- Speech Recognition
- E-commerce Technology
Background:
- E-commerce live streaming in China is a significant retail channel.
- Hosts may use subtle speech "morphs" to evade moderation and make unsubstantiated claims, posing consumer risks.
Purpose of the Study:
- To introduce and evaluate the Live Auditory Morph Resolution (LiveAMR) task.
- To develop a framework for restoring morphed speech transcriptions to their true forms.
Main Methods:
- Propose an LLM-enhanced training framework leveraging explanation knowledge (morph-type labels, LLM corrections, rationales).
- Mine knowledge from a frozen large language model without fine-tuning it.
- Fine-tune a lightweight T5 model using concatenated annotations and original sentences.
Main Results:
- Achieve substantial gains over baselines on both in-domain and out-of-domain test sets.
- Improve F0.5 score by up to 7 pp in-domain (to 0.943) and 5 pp out-of-domain (to 0.799) compared to a T5 baseline.
- Demonstrate efficient and accurate morph resolution using structured LLM-derived signals.
Conclusions:
- Structured LLM-derived signals can be effectively mined and injected into smaller models.
- The proposed method offers an efficient and accurate solution for auditory morph resolution in e-commerce live streams.
- This approach enhances consumer protection by improving the detection of misleading claims.
