Addressing the unique linguistic and cultural challenges of Chinese personality detection, this paper proposes a Dynamic Heterogeneous Graph Neural Network (DHGNN) model. DHGNN constructs contextualized post, psycholinguistic-enhanced, and cultural category nodes to better model polysemy, cultural specificity, and syntactic dynamism. It employs a dual-path message passing mechanism for cultural correlations and psychological salience, alongside a dynamic gated fusion module to adaptively integrate multi-path semantic features. Experiments on the Chinese Multi-label Affective Computing Dataset (CMACD) show significant improvements in Myers-Briggs Type Indicator (MBTI) classification. The work validates dynamic graph structures and cultural-psychological dual-path learning for disambiguation and feature enhancement, advancing Chinese personality detection methodology and providing foundations for applications like social media personalization and mental health monitoring. Future research will extend to multimodal fusion and cross-cultural transfer learning.