Instant Topic Extraction From a Text-Based Communication Channel for Seeing the World

This article proposes three topic extraction methods for a text-based communication channel of a message stream in bulletin board or chat services. A message is input into the channel, and these methods instantly select active and curious topics from the channel by using noun phrases, topic pressure at the latest message in the channel, topic unexpectedness, and partial matching. This serves as a module of a system that enables a user to follow unfolding world developments. An evaluative comparison of the performance of our methods and a conventional method using four data sets from two standpoints was performed. This is the first step in testing the performance of our methods.