Automated text clustering system on responses to open-ended questions in course evaluations

The challenge of this work is to provide an automated analysis model on responses to open-ended questions for course evaluations. As the first step, an automated text clustering system is developed to code individual responses into several categories. The categories are titled by keywords used in the responses and frequencies related to categories are given. A natural language processing algorithm considering word synonymy is used to measure similarity between two responses and applied first to classify responses to open-ended questions. Morphological parsing is performed for each response and concept dictionaries are used to calculate similarity between two words involved in responses. The sample responses written in Japanese over the 2001-2004 academic years in the computer literacy course were classified by the system and it is clarified that students tend to focus their responses on two special topics. Excellent classification accuracy is obtained.