Filtering method for chat logs toward construction of chat robot

Abstract

In recent years, the aging population has become one the major social issue. Elderly people living alone are social isolation and have few opportunities for conversations. To increase opportunities for conversation by elderly people living alone, a chat robot system which reuses the user utterance to robots' utterance have been developing. The problem of the previous system is that a system manager selects utterances unnecessary words from the utterances, revising of crude indecent language. Moreover, whether the utterance is suitable to be utilized as robot's utterance is judged manually which is a burden and considered a problem. In order to reduce the burden, this paper proposed two filtering methods. The first filtering method examined a simple pattern matching method. The second filtering examined three kinds of filtering method. We confirmed that the burden is reduced by using two filtering methods in combination. This study is positioned as the initial experiment against the problem of reducing the burden.