In recent years, most of text classification researches have used a term-based feature approach, but it has problems that those are language dependent and require a large number of data-set for analysis and learning the classification model. Also, in order to use feature language selection method, frequency ration per category of text classification has to be analogous to one another at the same time at least one feature language must exist among selected feature language in the expected text. A SNS(Social Network Service) message consists of 140 or less words. Therefore, it is very difficult to secure classification accuracy. On the other hand, if researchers use only language dependent feature, so called Meta-Data in the SNS message, it is very difficult to accurately classify SNS message which is reflected user's personal writing habits such as generation time, length, frequency in use of special symbols. And another problem is that the content in the SNS message is not reflected in the classification process at all. This study proposes a SNS message type classification system combining features that can be used in message for type classification in SNS environments and verifies.