Minority opinions abstraction

To deal with the multi-granularity web documents, we propose EMTF (EM Topic Fusion) algorithm to rank the by-joint possibility, resulting of their relevance to be the topic and their granularity. We first extract features from the different granularity web documents to establish the quantitative relationship amongst them. Then, the process of multi-granularity web documents analysis that leads to heretofore unknown information and opinions that valuable potential, minority and contentious respectively, which integrates the time, content, reprint and link information. Experiments show that EMTF achieves the best overall performance with high effectiveness and robustness.