Improving LLM Classification of Logical Errors by Integrating Error Relationship into Prompts