Students in network-based learning environments may have their own learning paths based on either their learning results or status. The learning system can choose suitable learning materials for individual students depending on students' learning results. There is a lot of research about learning diagnosis in distance education, and the main objective is to improve students' learning effects. This research proposes a full-loop learning architecture based on a knowledge map and provides feedback about teaching materials suitable for students. First of all, the learning system diagnoses and identifies the misconceptions of students by using a knowledge map; second, it selects suitable learning materials according to misconceptions and arranges a learning path for individual students to do remedial learning. This research uses precision, recall, and F-measure to measure the feedback effects. The results of the experiment show that the learning materials and learning paths suggested by the system are good. The contributions of this research are as follows: improving the diagnosis method; giving suitable learning materials and learning paths for remedy learning; and, moreover, improving the learning effects of students. (Contains 3 tables and 9 figures.)