||With the wide application and diverse access equipment of online community in recent years, people can easily browse and respond to online information, including photography, text, video and etc. However, a large part of text response, especially in micro-blogs, contains no essential meanings except the speaker. Half or more of them are unmeaning words such as greeting, advertising, murmurs, and irrelative arguments. Therefore, people cannot catch the main point of response in short time, and the subject will no longer be concerned and discussed.
For improving this, this study proposes an evaluate system for comments on Facebook’s wall posts, and ranking the importance of them. It can help user reading comments with high ranks directly and improve the quality of comments follow-up.
In this study, we classify all comments into 2 sets: (1) relative and (2) irrelative on every stage of : (1) Semantic analysis, (2) Meta-data features compare, and (3) Corpus similarity computing. Base on the result, we get True Positive Rate (TPR) as high as 0.9 for False Positive Rate (FPR) = 0.1.