Latent Dirichlet Allocation for textual student feedback analysis

• Main Authors: GOTTIPATI, Swapna, SHANKARARAMAN, Venky, LIN, Jeff
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2018
• Subjects: Teaching evaluations; Quantitative feedback analysis tool; Topic extraction; Sentiment Mining; Latent Dirichlet Models; Classification; Databases and Information Systems; Numerical Analysis and Scientific Computing
• Online Access: https://ink.library.smu.edu.sg/sis_research/4215; https://ink.library.smu.edu.sg/context/sis_research/article/5218/viewcontent/C3_01.pdf
• Institution: Singapore Management University

Education institutions collect feedback from students upon course completion and analyse it to improve curriculum design, delivery methodology and students' learning experience. A large part of feedback comes in the form textual comments, which pose a challenge in quantifying and deriving insights. In this paper, we present a novel approach of the Latent Dirichlet Allocation (LDA) model to address this difficulty in handling textual student feedback. The analysis of quantitative part of student feedback provides generalratings and helps to identify aspects of the teaching that are successful and those that can improve. The reasons for the failure or success, however, can only be deduced by analysing the textual comments from the students. In order to fully decipher the qualitative, textual feedback effectively and efficiently, researchers have attempted text mining techniques,which use natural language processing and machine learning algorithms to parse the text and extract the relevant insights. Our solution, using LDA models to discover the aspects or topics of the comments. We then employ sentiment mining techniques to classify the comments as positive or negative. To assess its performance, we applied our solution model on the data collected from teaching evaluations of Singapore Management University. Our experiments and evaluations show that LDA models perform better than clustering models in finding aspects from students' comments. In addition, the sentiment mining results indicate that classification method performs better than lexicon models. Also described in paper is the technical architecture of the tool along with some visuals of the interactive dashboard.