Topic evolution analysis can deeply identify and track the trend of changes in hot topics, which helps supply entire route of the topic evolution and reasonable advice to network public opinion monitoring. OLDA topic model is a commonly used tool for topic evolution analysis. But it has problems of old and new topics mixing and massive redundant words. Considering these problems, this paper proposes a dual-OLDA model by improving the genetic degrees of document-topic and topic-word distributions in OLDA model. Besides, a new method is put forward to improve topic-word distribution probability calculation. After testing on Chinese texts collected in Tianya forum, the experimental results show that the dual-OLDA model can detect new topics more easily and clearly explain the meanings of detected topics with lower probabilities of redundant words under Chinese semantic environment, which contributes to the evolution analysis of topics.