Adaptive knowledge graph construction and explainable reasoning algorithms for precise university teaching quality evaluation

The current problems in university teaching evaluation include a reliance on single evaluation methods, an inability to promptly identify issues in the learning process, and a situation in which only scores are known but not their underlying causes. To address these issues, this work proposes a precise teaching evaluation method based on the Adaptive Knowledge Graph (AKG). This method first integrates various data sources, such as student information, course structure, and learning behaviors, from the Open University Learning Analytics Dataset (OULAD). It then constructs a knowledge graph that can be continuously updated during the teaching process. Unlike traditional static knowledge graphs, the AKG can reflect changes in students’ learning status in real time. Subsequently, a hybrid reasoning mechanism combining the Graph Attention Network (GAT) and logical rules is designed. This mechanism can accurately predict students’ learning risks and provide clear bases for judgment. For example, a student who submits homework late and has few interactions is identified as being at high risk. Experimental results show that the accuracy of this method for the risk warning task reaches 91.2%, which is significantly superior to existing models such as the Knowledge Graph-Bidirectional Encoder Representations from Transformers (KG-BERT). Manual evaluation shows that 88.5% of the reasoning paths have clear teaching guidance value, and that evaluation errors between different student groups are less than 0.02. This work provides a new technical approach for building a transparent, fair, and reusable intelligent education evaluation system.