Knowlearn: Evaluating Cross-Subjects Interactive Learning by Deploying Knowledge Graph and Large Language Model
Poster Presentation Time: 1225-1400; 1500-1600
Venue: G4, Tai Po-Shek-O Room, Lower Level I
Presenter(s)
– Dr Hung-lin CHI, Associate Professor, Department of Building and Real Estate, The Hong Kong Polytechnic University
– Ms Junyu CHEN, Ph.D Student, Department of Building and Real Estate, The Hong Kong Polytechnic University
– Mr Haolei LIN, Ph.D Student, Department of Building and Real Estate, The Hong Kong Polytechnic University
Abstract
KnowLearn is an interactive learning assistant system designed for architecture, engineering, and construction (AEC) education, where personalized recommendations for students in virtual learning environments remain under-explored. An educational knowledge graph (KG) was constructed to contain multifaceted information by connecting pedagogical, learning performance, and learning feedback data as sub-graphs. A heterogeneous graph attention network (HAN) was implemented to infer latent information in the educational KG and identified essential factors shaping students’ acceptance of virtual learning environments. Based on sampling data of 107 students from the Hong Kong Polytechnic University, Department of Building and Real Estate, we found students’ self-efficacy, intention to use, and in-class quiz performance were significant predictors of final learning outcomes in subjects that adopt virtual learning environments. This project further deployed a local-based large language model (LLM) Qwen-7B and built an interactive graphical user interface (GUI) with Gradio. Utilizing the information preserved in the educational KG and learned from HAN as the basis, this LLM facilitated conversations between students and KnowLearn, enhancing personalized recommendations while securing student privacy. The developed system contributed to helping improve the learning experiences and performances of AEC students within virtual learning environments.