The ChatGPT craze has swept the globe, with over a billion users in two months. However, the accuracy of its answers has been widely questioned. After months of research and development, CHU’s Library and Information Service Center has introduced ChuGPT, a new-generation AI campus customer service chatbot. This significantly enhances answer accuracy, allowing students and parents to ask questions freely as the new school year begins.

CHU’s Library and Information Service Center Director, Dr. Chiu-Jung Tseng, mentioned that in response to the overwhelming popularity of ChatGPT and to assist teachers and students in quickly resolving various difficulties, President Liu Wei-Chi hoped to leverage generative language models to develop a new generation of school chatbots. After tireless efforts from colleagues at the Library and Information Service Center, they trained a domain-specific knowledge model using school web pages and documents. This model now works in synergy with the large language model GPT-3.5-Turbo, ultimately resulting in the completion of ChuGPT.

Photo1 / CHU’s Library and Information Service Center has launched ChuGPT, a new generation AI campus customer service chatbot, after months of research and training

Dr. Chiu-Jung Tseng pointed out that traditional customer service chatbots rely on manually curating question-answer pairs for training, which is labor-intensive and time-consuming. Moreover, because they do not utilize large language models, they have a limited understanding of the semantic context of questions, leading to reduced response accuracy. On the other hand, the new generation of customer service chatbots is trained directly using existing web pages and various documents, eliminating the need for manual curation. They leverage powerful large language models to understand the intent of questions and then extract answers from their self-trained domain knowledge models. This approach significantly improves response accuracy and eliminates concerns about knowledge theft.

Dr. Chiu-Jung Tseng, who has been mentoring students in researching automated question-answering systems for over 20 years, possesses extensive experience in natural language processing. She has also led several industry-academic projects that use automated question-answering systems to help businesses swiftly resolve issues through effective knowledge management, earning recognition from enterprises. This time, she has taken her students on a special journey to develop ChuGPT, combining her past research achievements with generative language models to create a new generation of automated question-answering systems.

Dr. Chiu-Jung Tseng stated that this technology suits campus customer service and is applicable to corporate customer service and knowledge management. Patents have already been applied for, and in the future, it is expected to be widely used in various industries. This technology is anticipated to significantly help companies reduce their customer service labor costs, enhance customer satisfaction, and improve work efficiency.