Location semantics can expose mobile users' preferences and lifestyles, posing a significant privacy risk when utilizing location-based services. Users with different semantic locations typically have varying sensitivities and different privacy requirements. However, current research has not quantitatively studied how sensitivity affects location privacy protection. This paper proposes a semantic adaptive geo-indistinguishability mechanism (SAGEO) to quantify personalized location privacy. Then, we present a novel semantic curved distance-based mechanism to ensure the above SAGEO achieve differential privacy by adding random noise to users' locations. Simulation results show that the proposed mechanism provides a better balance between privacy protection and quality of service compared to existing benchmarks.