labrador: a tool for automated grading support in multi...
TRANSCRIPT
Labrador: A Tool for Automated GradingSupport in Multi-section Courses
Drexel University Programming Learning EXperience (DUPLEX)
Departments of Mathematics† and Computer Science‡
http://duplex.mcs.drexel.edu
Christopher D. Cera‡, Jeffrey L. Popyack‡, Bruce Char‡, Robert N. Lass‡,
Nira Herrmann†, Paul Zoski†, Aparna Nanjappa‡
May 19, 2003
Roadmap
• Introduction
• Problems and Solution Goals
• Labrador
• Discussion
Slide 2 of 56
The Duplex Project: An Overview
• Take advantage of advances in Information Technology to improveinstruction and reduce costs for computer programming courses
• Modular Structure
– Multiple Entry Points– Multiple Audiences– Multiple levels of knowledge (Bloom’s Taxonomy)
• Computer Supported Cooperative Work (CSCW) in student labs
• Online Services - Today’s Topic
Slide 3 of 56
Course Redesign
• Emphasis on online materials
• WebCT introduced for:
– Delivering online course materials– Chat/Discussion groups– Quizzes/Labs– Electronic submission, grading, and return of assignments
• Issues: WebCT Interface does not handle all course needs
• Solution: Labrador
Slide 4 of 56
Who Am I?
• First TA in MCS to experiment with WebCT in December 2000
• Migrated course content to WebCT from previous course website
• HTML assignments and labs into question database
• Gradebook maintainer
• Wrote demo’s and documentation to train other TAs
• Developer of software supplements to WebCT to support courseadministration
Slide 5 of 56
Roadmap
• Introduction
• Problems and Solution Goals
• Labrador
• Discussion
Slide 6 of 56
Before WebCT
• Hundreds of people were involvedin paper exchanges of handwrittenassignments and quizzes
• Testing programs required floppyexchanges
• No Chat
• No newsgroup-style threads
• Feedback and grades are notonline
Slide 7 of 56
WebCT
• Service from InformationResources and Technology (IRT)(since v3.1)
• MCS started Dec. 2000 (v3.5)
• Currently using version 3.8.3
Slide 8 of 56
Features
• General Course Website
• Centralized Administration
• Labs: Online Quizzes withAutomated Grading
• Homework: Online Assignments
• Joint Staff–Student Chat
• Discussion Threads
Slide 9 of 56
Large Classes and WebCT
• Even with WebCT .... still somedifficulties:
– Bulk download of assignmentfiles and quizzes for grading
– Handling select sectionsrequires searching and clicking
– Files submitted in compressed,archived, or encoded formatsare tedious to unpack manually
– Transferring data to othersystems
Slide 10 of 56
Software Design Goals
• Bulk Assignment Downloading (prior to
3.7)
• Bulk Quiz Downloading
• Section Sorting for bulk downloads, or
only one section
• Post-Processing student files
– Archive Extraction (tar, zip)
– Decompress (gz, zip)
– Decode (uue)
• Minimal staff intervention when
transferring submissions between
systems, eg. Plagiarism Detection
Systems
Slide 11 of 56
Software Design Goals [continued]
• Automatically collate source code
• Generate electronic documents tofacilitate grading and archiving
• Upload grades and marked-updocuments
• Remote Execution downloading tocomputer x (on campus) whileoperating at computer y (offcampus)
Slide 12 of 56
The Bigger Picture
• Course-Specific Tasks
• Not all processing should be doneon the server
• Select files have to be transferredto a different system for furtherprocessing and analysis by staff
• Client-side support is needed toperform this, preferablyautomated and not necessarilyusing a web browser.
Slide 13 of 56
Roadmap
• Introduction
• Problems and Solution Goals
• Labrador
• Discussion
Slide 14 of 56
Labrador: Our Solution
• Client-side WebCT Supplement
• Cross-platform
• Works for users with TA andDesigner access to WebCT
Slide 15 of 56
Labrador Execution Timeline
SubmissionDownloader
SectionSorting
Post-Processing
PDFGeneration
JPlag
Moss
SubmissionUploader
Slide 16 of 56
User Interface
• Different users / Different UIpreferences
• GUI
• Command-line
• Interactive
• Configuration File
Slide 17 of 56
Bulk Downloader
• Web-Crawler: simulates clicks ofactual staff member
• Parses HTML to find desired textor URLs to crawl to next
• Works on assignments and quizzes
• Only component in Labradorwhich interacts with WebCT
Slide 18 of 56
Required Information
http://webct.drexel.edu/SCRIPT/CS164_Fall2002/scripts/designer/dropbox_edit.pl?DROPBOX_ASSN_VIEW+_side_nav++1006285909
Username/Password st96k9ry
Server URL webct.drexel.edu
Course ID CS164 Fall2001
Submission Name or ID Recursion II or 1006285909
Optional Username List unames.txt
Slide 19 of 56
Organizing Submissions by Section
• Organizes each student’ssubmissions into a separate folderfor each section.
• How to tell Labrador the sections:
– Creating a Section column ingradebook
– “Username, Section” CSV File– Username file
Slide 20 of 56
PDF Generation for Electronic Mark-up
• Adobe Portable Document Formatis available on all major platforms
• With Adobe Acrobat, PDFs canbe annotated by graders
• Sony VAIO Slimtop PC(PCV-LX920)
• Toshiba Tablet PC: Protege 3500
• Acer Tablet PC
• Wacom Pen Tablet
Slide 21 of 56
Demonstration
Startup screen prompts for the username and password
Slide 24 of 56
Demonstration [continued]
TA enters username and password
Slide 25 of 56
Demonstration [continued]
Labrador prompts for the course name and optional student list
Slide 26 of 56
Demonstration [continued]
TA enters course name and student list
Slide 27 of 56
Demonstration [continued]
Labrador prompts for the submission type
Slide 28 of 56
Demonstration [continued]
TA selects assignments
Slide 29 of 56
Demonstration [continued]
TA selects post-processing
Slide 30 of 56
Demonstration [continued]
TA selects PDF generation
Slide 31 of 56
Demonstration [continued]
Labrador prompts for the specific assignment
Slide 32 of 56
Demonstration [continued]
TA selects “Practice Assignment” and begins downloading
Slide 33 of 56
Demonstration [continued]
Labrador notifies the TA that the job is complete
Slide 34 of 56
Demonstration [continued]
Exploded view of TA’s folder
Slide 35 of 56
Redistribution
• How can we return annotated PDF’s back to students using WebCT?
• Version 3.8 addresses this issue
• Labrador supports this upload feature
Slide 36 of 56
Labrador Applications
• Interface between WebCT documents and other software
– Decompressing files– Reformatting files for grading (PDF)– Submission to Plagiarism Detection Software (Moss/JPlag)– Other third party software programs– Returning processed/graded documents to WebCT
• Primary Issue: Compatibility with heterogeneous systems
Slide 37 of 56
Heterogeneous Systems
• Each system requires data to bepackaged in a different way
• Plagiarism Detection Systems
• Moss [1] and JPLAG [2] havebeen used extensively
• Other processes (e.g. PDFgenerator)
• Future work: Automatic programcompiling and testing
Slide 38 of 56
Automated Plagiarism Detection
• Digital formats make ”borrowing”easy
• Browsing similar works needs asimple and quick user interface.
• Careful review by faculty to assessresults and present to students
Slide 39 of 56
Moss [1]
• C, C++, Java, ML, Lisp, Scheme, Pascal, and Ada
• Common code feature reduces false positives
• http://www.cs.berkeley.edu/∼aiken/moss.html
Slide 40 of 56
MossCliques Interface
http://duplex.mcs.drexel.edu/software/MossCliques.zip
Slide 42 of 56
JPlag [2]
• C, C++, Scheme, and Java
• For plain text files, it matches a user specified number of wordsappearing in succession
• Could be used for any course grading written (text) documents
• http://wwwipd.ira.uka.de:2222/
Slide 43 of 56
Roadmap
• Introduction
• Problems and Solution Goals
• Labrador
• Discussion
Slide 45 of 56
Recent WebCT Enhancements
• Relevant to this talk:
– 3.7 Addressed bulk downloadissue for assignments
– 3.8 Attaching documents to anassignment
Slide 46 of 56
Future WebCT Enhancements
• Power users will need functionalitynot yet supported
• Every domain will also requireadditional functionality
• Not feasible for all domain-specificfunctionality to run on theWebCT server
Slide 47 of 56
HTTP: Insufficient For Data Interchange
• Was designed for visual content
• Heavy client interaction
• An HTTP based approach is sensitive to the exact location of webpages, and format of text within them
• One possible solution is ...
Slide 48 of 56
API for non-administrators
• Stateful protocol so clients can bebuilt by 3rd parties
• Could be coupled with efforts ofthe “semantic web” (e.g. RDF)
• This would eliminate the need forLabrador to “screen-scrape” textfrom web pages
Slide 49 of 56
Labrador Availability
• Contact Us
http://duplex.mcs.drexel.edu
Slide 50 of 56
The Group
Bruce Char Professor, Computer ScienceNira Herrmann Professor and Head, MathematicsJeffrey L. Popyack Associate Professor, Computer SciencePaul Zoski Instructor, Math and Computer Science
Christopher D. Cera Computer Science Graduate StudentRobert N. Lass Computer Science Undergraduate StudentAparna Nanjappa Computer Science Graduate StudentDerek Rosenzweig Computer Science Undergraduate StudentJasper Zhang Computer Science Undergraduate Student
Slide 51 of 56
Project Support
• National Science Foundation,
Division of Undergraduate
Education, DUE − 0089009
• The Pew Learning and
Technology Program at the
Center for Academic
Transformation
• The Ramsey-McCluskey Family
Foundation, Margaret Ramsey,′84
• Drexel University
Slide 52 of 56
References
[1] Alex Aiken. MOSS: A System for Detecting Software Plagiarism(Unpublished), http://www.cs.berkeley.edu/∼aiken/moss.html.
[2] L. Prechelt, G. Malpohl, and M. Philippsen. JPlag: Finding PlagiarismsAmong a Set of Programs. Technical Report 2000-1, Fakultat furInformatik, Universitat Karlsruhe, Germany, March 2000.
Slide 53 of 56
Related Work
[1] Robert N. Lass, Christopher D. Cera, Nathaniel T. Bomberger, Bruce Char, Jeffrey L.
Popyack, Nira Herrmann, and Paul Zoski. Tools and Techniques for Large Scale Grading
using Web-based Commercial Off-The-Shelf Software (To Appear). In Proceedings of
the Eighth Annual Innovation and Technology in Computer Science Education (ITiCSE).
ACM Press, June 2003.
[2] Nira Herrmann, Jeffrey L. Popyack, Bruce Char, Paul Zoski, Christopher D. Cera,
Robert N. Lass, and Aparna Nanjappa. Redesigning Computer Programming Using
Multi-level Online Modules for a Mixed Audience. In Proceedings of the Thirty-Fourth
SIGCSE Technical Symposium on Computer Science Education. ACM Press, February
2003.
Slide 54 of 56
Related Work [continued]
[3] Jeffrey L. Popyack, Nira Herrmann, Paul Zoski, Bruce Char, Christopher D. Cera, and
Robert N. Lass. Academic Dishonesty in a High-Tech Environment (Special Session). In
Proceedings of the Thirty-Fourth SIGCSE Technical Symposium on Computer Science
Education. ACM Press, February 2003.
[4] Jeffrey L. Popyack, Bruce Char, Paul Zoski, Nira Herrmann, Christopher D. Cera,
Robert N. Lass, and Aparna Nanjappa. Course Management Systems (Birds-of-a-
Feather Session). In Proceedings of the Thirty-Forth SIGCSE Technical Symposium on
Computer Science Education. ACM Press, February 2003.
Slide 55 of 56
Related Work [continued]
[5] Jeffrey L. Popyack, Bruce Char, Nira Herrmann, Paul Zoski, Christopher D. Cera, and
Robert N. Lass. Pen-Based Electronic Grading of Online Student Submissions. In
Syllabus fall2002, Technology for Higher Education Conference, November 2002.
[6] Christopher D. Cera, Robert N. Lass, Bruce Char, Jeffrey L. Popyack, Nira Herrmann,
and Paul Zoski. Labrador: A Tool for Automated Grading Support in Multi-Section
Courses. In Proceedings of the Fourth WebCT User Conference, Integrating the
Campus: Technical Solutions for Resource Development or Wide Scale E-Learning
Deployment, July 2002.
[7] Jeffrey L. Popyack, Bruce Char, Paul Zoski, Nira Herrmann, and Christopher D. Cera.
Managing Course Management Systems (Birds-of-a-Feather Session). In Proceedings
of the Thirty-Third SIGCSE Technical Symposium on Computer Science Education,
page 423. ACM Press, February 2002.
Slide 56 of 56