lisabecker / pm_textmining_teacherreportsclassification

Text mining project for the classification of "self reports" from teaching students (school project, 12 ECTS).

Classification of pre-service science teacher's self reflections

Text mining project for the classification of "self reports" from teaching students. Project Module for the Cognitive Systems Master's program of the University of Potsdam (12 ECTS). This project was in cooperation with the Physics Education Research, Institute of Astronomy and Physics of the University of Potsdam. Supervised by Professor Manfred Stede at the University of Potsdam.

The data is collected from pre-service science teachers and is under non-disclosure.

This project investigated whether natural language processing can be used to automate the process of analyzing self-reflection reports and giving automated feedback to the pre-service teachers on their enactment. Several Machine Learning algorithms were exploited, amongst others Naive Bayes, Linear Support Vector Machines, Logistic Regression and Random Forest Classifiers. Different features were compared to find the optimal combination for classification. The Random Forest Classifier with the positioning of of sentences in a given document proved to be the best approach with a classification accuracy of 76,6% (compared to 72% of inner-annotator accuracy).

The paper "Utilizing Natural Language Processing to Analyze Preservice Science Teachers' Reflections of Their Teaching Enactments" got published in April 2020.