The Use of Filtered Linear Discriminant Analysis for Automatic Discussion Rating

• In Event: Poster Session 4
  In Poster Session: Advanced Technologies for Learning SIG Poster 2

Abstract

One challenge of utilizing video discussion in education is assessing its quality. Similar to automatic essay scoring, automatic discussion scoring provides a feasible solution to large scale discussion scoring. In any application of automatic scoring, a classification algorithm has to be specified. Linear Discriminant Analysis (LDA) as a classification algorithm is computationally fast. However, lacking of tuning parameter makes it prone to over-fitting. To address this issue, we propose a filtered LDA algorithm that uses Principle Component Analysis (PCA). We demonstrated effectiveness of the algorithm by classifying a real video discussion dataset and compared the classification results with those from the traditional LDA.

Authors

• Xiang Liu, Teachers College, Columbia University

• Zhou Zhou, Teachers College, Columbia University

• Hui Soo Chae, Teachers College, Columbia University

• Gary J. Natriello, Teachers College, Columbia University