Hot topic extraction in microblog has received great interests during recent years. Traditionally, there are two different frameworks for this task, one being based on supervised learning, and the other being based on unsupervised learning, which typically exploits clustering or topic models for the task. Actually, the construction of large-scaled annotated corpus for supervised learning is extremely difficult, thus most work exploits unsupervised learning. In this paper, unsupervised learning is used. Different from previous work, the hashtags labeled by microblog authors are used as the key information to improve the performance of hot topic extraction. Finally, the experiment demonstrate that this information is highly useful for hot topic extraction in microblog.