Microblog users generally publish their opinions by using condensed text with some non-textual content. Besides, post responses from participants often include noise such as chaotic messages or unrelated information to the theme. Thus, we propose a Feature-based Filtering Model attempts to filter these noises. Moreover, we propose a method, which select the responses based on user participation degree, Maximum Discussion Group Detection (MDGD), to solve the problem of ignored information by current content fusion approaches. Briefly, the posts with higher user participation degree are selected to extract the short text from original post and its responses. The related content from several microblog platforms is also referred to enrich the fusion results. In the experiments, the test data set is collected from the microblog platforms of Plurk and Facebook. Finally, the Normalized Discounted Cumulative Gain (NDCG) metrics show that our method is capable to provide qualified extraction results.