The knowledge graph, with its continually expanding dataset, has found diverse applications across domains such as product recommendations and knowledge question answering. However, the presence of erroneous and conflicting information resulting from automated mechanisms and crowd-sourcing poses a considerable challenge to the integrity and veracity of the knowledge graph. To tackle this problem and the lack of interpretability in existing trustworthiness measurement methods, we propose an innovative approach to triple trustworthiness measurement for knowledge graphs, leveraging reinforcement learning. Our contribution lies in the design of a noise-tolerant knowledge embedding, which exploits reinforcement learning for triple selection and trustworthiness measurement. Our experiments evince the effectiveness of this approach in detecting noisy triples and providing evidence for triple measurement.