Date on Master's Thesis/Doctoral Dissertation

5-2011

Document Type

Master's Thesis

Degree Name

M. Eng.

Department

Computer Engineering and Computer Science

Committee Chair

Desoky, Ahmed H.

Subject

Grading and marking (Students)--Data processing; Discourse analysis--Data processing; Instructional systems--Design

Abstract

In recent years the use of Tablet PCs in education has received a great deal of attention. Improved note-taking and lecture presentation have been the main focus of Tablet PCs in education. Currently, grading of student work collected as digital ink using Tablet PCs is still done mostly as it would be done if collected as pencil on paper. However, having content stored as digital ink provides an opportunity to perform analysis that is neither practical nor possible with pen and paper student content. Once student work is collected using DyKnow Vision, an interactive Tablet PC based classroom software tool, the student names and specific answers to questions can be extracted for analysis. One type of analysis is clustering of students’ answers. For short answer English words, clustering of answers can be automated using handwriting recognition algorithms, existing clustering techniques and string distance calculations. The clustering of answers will be an automated process that forms sets, but could be supplemented with human feedback to further refine the result. A software system to implement this approach was designed and developed. Multiple string distance measures were used to implement an agglomerative clustering algorithm that brought together similar answers. Initial evaluation of this approach on short, English word, answers showed that student answers can be clustered in such a way that produces useful results for a human grader. The algorithm has been found to be useful at creating groups of answers which a grader would consider to be identical with the most logical merges occurring early. However, the heuristic used by the algorithm has been found to both stop grouping similar answers too early in some cases and too late in others. Further refinement of this heuristic is needed to produce ideal clusters. While human processing is still required, more development in this area and the use of more advanced techniques would be highly valuable as technology becomes more tightly integrated in the classroom.

Download

Share

COinS