Abstract: Topic modeling is one of the most important text mining methods, which has been widely used in analyzing the topic composition of a text corpus. Its main drawback lies in that it is difficult to interpret or adjust the topic modeling results. To help users understand and manipulate topic models, we design and implement a progressive visual analysis framework with two visualization components: a corpus refinement component which assists users construct the dictionary efficiently;and a topic modelling component which illustrates multi-dimensional information concerning topics and allows for interactive manipulation of topic models. The effectiveness of the proposed approach is tested with a control experiment using the 20 newsgroups news dataset. A case study on the real Douban movie dataset further verifies the practicability of TMvis.