In this study, we developed a classification system of programming errors based on the historical data of 680,540 programming records collected on the Online Judge platform. The classification system described six types of programming errors (i.e., syntax, logical, type, writing, misunderstanding, and runtime errors) and their connections with fundamental programming knowledge. Furthermore, we used ontology-based learner modeling techniques to create student ontology, which provided an accurate representation of a student's information (e.g., knowledge level, programming history, and performance) and the mechanisms for tracking its continuous changes. We also designed problem ontology, providing a uniform approach to describe the characteristics of a programming problem. The instances of student and problem ontologies were visualized as knowledge graphs. Based on the classification system of programming errors and knowledge graphs, we designed an automated system for assigning programming tasks to students. We tested the effectiveness of the automated task assignment system using a quasi-experimental design. Students in the control group were asked to solve programming tasks assigned by their teacher throughout eight weeks. In the experimental group, students accomplished programming tasks assigned by the system. We found no significant difference in student performance between the two groups. This study has significant methodological and practical implications.