Automatic labeling method of utterance intention for werewolf BBS using self-validation mechanism

Abstract

We propose an automatic labeling method of utterance intention using key words in the topic model for werewolf BBS using self-validation mechanism. At first the topic model is used to assign labels by considering the topic ratio of the utterance and the important words in the utterance. Then, the support vector machine and the decision tree analysis are used to evaluate current assignment and reassign the labels, which is a repetitive process working as a self-validation mechanism and finaly induces adequate label assignment. Experimental results using a certain werewolf BBS show effectiveness of the proposed method.