Page 316 - AI for Good Innovate for Impact
P. 316

AI for Good Innovate for Impact



                          Use Case 6: The "CloudTop": AI-driven Marketing and Service

                      Intelligentized Applications of China Telecom Cloud









                      Organization: China Telecom Cloud Technology Co., Ltd
                      Country: China


                      Contact Person(s):
                      Aijiao Wang, wangaijiao@ chinatelecom .cn
                      Qifang Liu, liuqf2@ chinatelecom .cn
                      Wenwei Li, liww9@ chinatelecom .cn


                      1      Use Case Summary Table

                       Item              Details

                       Category          5G
                                         There is an excessive amount of low-level, repetitive work in simple scenar-
                                         ios, and some of these tasks have a negative impact on employees' physical
                       Problem           and mental health, as well as their career development. The Conventional
                       Addressed
                                         digitalization solutions often struggle to support the complex needs of
                                         China's industries in their digitalization and intelligence transformation.

                                         Develop a vertical domain model tailored to the enterprise by integrating
                                         external enterprise knowledge into a general-purpose large language
                                         model. Implement a Retrieval-Augmented Generation (RAG) system that
                                         leverages high-quality external knowledge bases to enhance the accu-
                                         racy and reliability of responses. Utilize Automatic Speech Recognition
                       Key Aspects of  (ASR) models to convert speech into text, enabling precise recognition
                       Solution          of customer intent. Apply efficient semantic retrieval techniques, such as
                                         dense vector retrieval, and integrate multiple models to balance speed and
                                         accuracy for tasks including question summarization, sentiment analysis,
                                         entity extraction, and product categorization. Additionally, build self-re-
                                         flective agents capable of generating self-feedback to optimize their
                                         decision-making logic over time.

                                         Natural Language Processing (NLP), Large Language Models (LLMs), Text
                       Technology        Classification, Data Analytics, Fine-tuning, Retrieval-Augmented Genera-
                       Keywords          tion (RAG), Prompt Engineering, ASR, Sentiment Analysis, Self-reflection
                                         Agent

                                         The project uses both public and private data [2]. The knowledge docu-
                       Data Availability  ments part of the model training data are accessible from the China
                                         Telecom Cloud official documentation webpage [8]. 

                                         Audio and voice data, text data, image data, numerical data, and label
                       Metadata (Type of   data are generated during customer service interactions and enterprise
                       Data)
                                         solution delivery.







                  280
   311   312   313   314   315   316   317   318   319   320   321