differential PLSA

A Method of Extracting not Representative Topics but More Individual Topics from Text Data

Abstract

This study proposes a new method extracting topics from text data named differential PLSA. It enables to extract not representative topics but more individual ones. This paper showed the effectiveness by applying the method to patent document data and comparing with results using normal PLSA. As a result, topics extracted by the method were composed of less frequent and more concrete elements, and they were more individual.