The document presents a new online semantic-enhanced Dirichlet model (OSDM) for clustering short text streams. OSDM addresses challenges with existing approaches like semantic ambiguity, concept drift over time, and batch vs online processing. It maintains active topics online using a non-parametric probabilistic graphical model that incorporates semantic information through term co-occurrence and performs automatic topic detection. Experimental results on news, tweet and Reuters datasets show OSDM outperforms other models in clustering performance over data streams and is robust to parameter changes.