CodeSimilarity: an approach for clustering introductory programming assignments

Osei-Owusu, Jonathan

Permalink

https://hdl.handle.net/2142/108644 Copy

Description

Title

CodeSimilarity: an approach for clustering introductory programming assignments

Author(s)

Osei-Owusu, Jonathan

Issue Date

2020-07-24

Director of Research (if dissertation) or Advisor (if thesis)

Xie, Tao

Department of Study

Computer Science

Discipline

Computer Science

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• clustering
• programming education
• testing

Abstract

Enrollment in introductory programming (CS1) courses continues to surge and hundreds of CS1 students can produce thousands of submissions for a single problem, all requiring timely feedback and accurate grading. While not exclusive to CS1 courses, instructors of such courses are challenged to provide feedback at scale (e.g., to hundreds of students). Because these students have a diverse range of skills and backgrounds, it is essential to differentiate common strategies and shortcomings of student submissions to a given problem. There is a strong need for clustering submissions by the similarity of their strategies for enabling instructors to provide customized feedback to students. To fill this need, in this thesis, we present the CodeSimilarity approach, which first automatically generates test data for correct student submissions and then uses semantic program features (i.e., path conditions) to cluster correct student submissions by their strategies. We define the strategy employed by a student submission as the way that the problem space is partitioned into sub-spaces and how the problem is uniquely addressed within each sub-space. In particular, CodeSimilarity leverages automated test generation based on symbolic execution to determine the path conditions for a given submission; comparing each submission’s path conditions allows to establish behavioral equivalence relationships with respect to the strategies employed by these submissions. We evaluate CodeSimilarity on four datasets to assess the effectiveness of our approach. The evaluation results show that by using semantic program features (i.e., path conditions), CodeSimilarity can effectively cluster submissions that employ the same strategy.

Graduation Semester

2020-08

Type of Resource

Thesis

Permalink

http://hdl.handle.net/2142/108644

Copyright and License Information

Copyright 2020 Jonathan Osei-Owusu

Owning Collections