A Study of Network Informal Language Using Minimal Supervision Approach

The subjective text is an important processed object of opinion mining, but sometimes there have been many informal expressions in a subjective text. The authors of the subjective texts have the personal expression habits which are not restricted to a formal grammar, so Network Informal Language (NIL) emerges. For example, the formal word “ (like)” is usually replaced by the NIL word “ tlx (conjee)” in some chatting tools like OICQ (Open ICQ) and MSN (Microsoft Ser vice Network). Currently, opinion mining tools are less effective to dispose these problems. For instance, they regard the NIL word “8 fa (non-word)” as a text noise, which expresses one's viewpoint, that is, the formal word “ (ok)” in fact. In this paper, we propose an approach to try to resolve the problems from NIL expres sions. Because NIL expressions are manually identified in general, we use minimal supervision technology to identify them. According to the two different types of NIL, we adopt different strategies for those. The experiment results have shown that the performance of the proposed approach is satisfied. Therefore, this approach is reasonable and effective. In the future, we will improve this approach, so that it can process more complicated NIL expressions.