With the rapid development of web2.0, there is more and more content on social media, and information is widely spread in people’s lives through social media. Public often make vast opinions on hot Events on social media platforms, such as Sina Weibo and Twitter. Clustering these opinions can increase understanding of the semantics of public opinions. Mining these opinions thoroughly can help companies and management make better decisions. The challenge of opinion clustering for hot events is that most of opinions contain background information of event. The background information could reduce opinion clustering performance. In this paper, we propose a topic model named background removal LDA(BR-LDA) model for opinion clustering. The model adds the idea of removing background to the LDA model so it can separate opinion words from background words. First, we remove some words with high frequency in the corpus. Then the model applies BR-LDA model to automatically cluster public opinions. Experimental results on two real-world datasets of two languages, Chinese and English, verify the efficiency of the proposed model.