Page 324 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

Use Case 7: “Di Ting” Customer Service Agent

Organization: China Telecom Co., Ltd. Hubei Branch

Country: China

Contact Person(s):
Wang Wenduo, 18907178900@189.cn
Jiao Hailong,18907100678@189.cn
Zhou Huiping, hbzhouhp@ chinatelecom .cn

1 Use Case Summary Table

Item Details

Category 5G
Traditional natural language understanding (NLP) technology is difficult to
handle increasingly complex customer service business scenarios. In prac-
Problem
Addressed tical applications, a significant amount of human intervention is required as
a fallback, leading to issues such as inconsistent service quality, repetitive
tasks, low service efficiency, and high operational costs for enterprises.

Introducing the AGENT architecture to build the "Diting" intelligent
customer service agent: Using Large Language Model (LLM) for intent
recognition, entity recognition, sentiment analysis, dialogue analysis,
Key Aspects of business opportunity mining, and more. Leveraging the knowledge base
Solution
to store customer service marketing activity data, telecommunication
business knowledge data, and other relevant information for product
recommendations.

Technology Large and small model collaboration, Intelligent Customer Service
Keywords

Data Availability Not Available

Metadata (Type of
Data) Text data, audio data

Fully parameterize the training of large and small models for intent recog-
nition, entity recognition, and incorporate decision-making judgments
to enhance the collaborative capabilities of large and small models. The
Model Training and foundational small model used for this task is Tiny-BERT [4] and the large
Fine-Tuning model used for intent recognition is the Telechat32B [5] or Qwen8B [6].
These are fine-tuned on training data that consists of dialogues between
users and customer service agents.Inference-based large models are used
for the automatic optimization and generation of service solutions.

288