Enhancing Semantic Consistency of Large Language Models through Model Editing: An Interpretability-Oriented Approach