Noise Filtering in Chat Data Based on Social Network

Noise brings great troubles in inspecting chat room because it reduces the monitoring efficiency of chat room. But there are no otherclues to distinguish noise and normal utterances directly except time clue, it is difficult to filter noise. With the temporal relations of dialogs, thispaper proposes a new method to handle the problem. It deduces the social network by the temporal relations of dialogs, and then judges the noise andfilteres in terms of the features of chatters’ relations expressed by social networks. Experimental result shows that the method can judge noise welland improve the accuracy of topic detection.