A Method of Labeling to the Topic in a Chat Dialogue

In this research, we propose a method of labeling topic word for the sections in chat dialogue. First, investigated how humans label words. As result, it was found that words included first sentence which started and associated with appearing evenly section were given as It also non-characteristic appear other selected when is based on frequency appearance section. We obtain high TF-IDF candidate word. The DF value regardless are set to exclude stop compared using proposed labels by humans. confirmed accuracy improved.