Page 317 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Models: ASR is powered by China Telecom Cloud's self-developed ASR
technology. NLP uses TeleChat2 and Huize (self-developed large language 4.3 - 5G
models by China Telecom), DeepSeek, and BGE.
Training and Fine-Tuning: The ASR model is trained on annotated customer
voice data. NLP models are fine-tuned with text, image, and numerical data
Model Training and to develop a vertical domain model capable of answering cloud indus-
Fine-Tuning try-specific queries. By incorporating annotated data for specific tasks,
we guide the model to optimize for those tasks. For example, we train a
classification model for customer service scenarios using large amounts of
labeled classification data, which helps Customer Service Representatives
automatically tag product categories and issue categories for customer
service requests.
The use case is part of a larger product development; It’s now in produc-
Testbeds or Pilot
Deployments tion at China Telecom Cloud. Services have been deployed through the
Ctyun portal [7]

Code Repositories Not Available

2 Use Case Description

2�1 Description

Traditional marketing and service models are labour-intensive, with a lot of repetitive, low-skilled
work in simple scenarios. This leads to high turnover rates and doesn't support employees'
well-being or long-term career development. At the same time, one-size-fits-all digital solutions
often fall short in supporting industry-wide digital and intelligent transformation.

To address the above industry challenges, this case proposes an advanced solution. Based on
leading frontier AI technologies such as NLP, LLM, Artificial Intelligence Generated Content
(AIGC), RAG, Base General Embedding (BGE), Prompt Engineering, ASR, and Sentiment
Analysis, it upgrades the technologies and working modes for tasks like intent understanding,
knowledge retrieval, intelligent question answering, solution recommendation, and risk
monitoring. It has realized intelligent marketing service application capabilities such as intelligent
robots, intelligent knowledge bases, intelligent recommendations, intelligent assistance, and
risk monitoring. It is committed to creating emerging technology positions, enhancing labour
productivity and employee satisfaction, providing demand insights and customized solutions,
promoting rational consumption of digital services, accelerating the digital transformation of
enterprises, and achieving high-level digital economic growth.
Consider the scenario of using intelligent assistance to reduce the low-level repetitive work of
service personnel. This case fully exploits the value of diverse data. By using text data, image
data, and numerical data to conduct fine-grained fine-tuning of the model, it helps the model
complete the training process, turning it into a vertical domain model that is deeply adapted
to the needs of specific industries and can accurately recommend knowledge in professional
fields.

During the actual service process carried out by service personnel, the system uses ASR
technology to convert customers' voice data into text content and extract questions.

281