It is important for teachers to understand whether students have learned correct concepts or misconceptions. To identify correct concepts and misconceptions, we attempted to analyze the characteristics of students' concept descriptions using machine learning. It has been mentioned that university students retain the misconception that 'light refracts at the center line of a convex lens' due to the influence of the learning construction method in junior high school. Therefore, the descriptions from 104 Japanese university third-graders (36 from the faculty of social studies and 68 from the faculty of science and engineering) were analyzed by text mining using a support vector machine, and their feature words were identified. As a result, it was possible to classify the characteristics of the descriptions as a correct concept or misconception based on the feature words.