This work presents SIKG, a semantic-aware test selection and prioritization technique that combines fine-grained change classification with graph-based impact propagation. Our approach employs reinforcement learning to continuously adapt the knowledge graph weights and test selection policies based on empirical feedback from test execution results. We implemented our technique in a Visual Studio Code extension that integrates seamlessly with existing development workflows. Our evaluation of seven diverse Python projects demonstrates that SIKG achieves an average test suite reduction of 74.4% while maintaining a 78.2% fault detection rate (at a 60% reduction). Our approach outperforms traditional approaches and achieves 20.8% higher precision than structural-only approaches through semantic understanding of change impact. The implementation of our approach is available publicly at https://github.com/ai4se4ai-lab/SIKG.