Knowledge-oriented Maintenance at the University of Ottawa

1
Knowledge-oriented Maintenanceat the University
of Ottawa

• Timothy C Lethbridge
• KOM 2004 - Banff

2
The Knowledge-Based Reverse Engineering Group at
the University of Ottawa

• Formed by Tim Lethbridge in 1994
• Original focus was strictly knowledge-based work
• We still do some of that
• But our focus has broadened to tools for helping
  people deal with complexity

3
Project 1 Bullding a Knowledge Base to help
software engineering of ObjecTime

• History
• We had built a series of knowledge management
  systems (CODE2, CODE4) in the 1989-1994 period
• We were working with Nortel to help them grapple
  with the complexity in their software design
  tools
• We built a KB interactively with their developers
• The process helped iron out key concepts
• Later, this tool span off as a separate company
• Incorporated into Rational, and later into IBM
• Reference
• My PhD thesis http//www.site.uottawa.ca/tcl/the
  sis_html/thesis_ToC.html

4
Project 2 A lightweight knowledge base for
maintainers

• Contained
• An ontology of concepts relating to a large
  telecom system at Mitel
• Basic definitions and cross references
• Uses
• Available for exploration to help new hires learn
  the system
• Hooked to intelligent search system to help
  search for concepts in source code
• Hooked to browsing system When browsing code,
  maintainers could follow links from words in
  comments

5
Project 2 - continued

• Experience building the KB
• Done quickly with very little work
• Step 1 Scan source code comments and key
  documents for words and phrases
• Step 2 Use a multi-phase ranking/voting UI to
  have project team members prioritize, organize
  and categorize concepts
• Deployed
• In TkSee at Mitel - used by 15-20 maintainers

6
Project 2 - continued

• References
• Sayyad-Shirabad, J., Lethbridge, T.C. and Lyon,
  S, (1997, May), "A Little Knowledge Can Go a Long
  Way Towards Program Understanding", 5th
  International Workshop on Program Comprehension,
  Dearborn, MI, pp. 111-117
• Liu, H., and Lethbridge, T.C. (2002),
  Intelligent Search Methods for Software
  Maintenance, Information Systems Frontiers, 4,
  4, pp. 409-423.

7
Project 3 Mining a maintenance records to find
relevant software entities

• Premise
• When maintaining software, maintainers are often
  not aware of the files, classes, methods that
  need to be modified
• We want to create an advisor that will study what
  you are modifying, and suggest other things that
  might need modifying
• This can be done by systematic impact analysis,
  but we want to try an alternative approach

8
Project 3 - continued

• Approach
• Extract all the software entities that have been
  modified together over several years in the
  system
• Found in the change management system
• Extract a wide variety of attributes of all
  software entities
• Apply machine learning to derive a model that
  predicts, for any pair, whether when modifying
  one element, you should look at the other
• Results
• After a major PhD project, we have very good
  prediction tools

9
Project 3 - continued

• References
• Sayyad Shirabad, S, Lethbridge, T.C. and Matwin,
  S. (2004) Mining Software the Repository of a
  Legacy Telephony System, MSR 2004 International
  Workshop on Mining Software Repositories, in
  conjunction with ICSE 2004, Edinburgh Scotland,
  May, IEE Press
• Sayyad Shirabad, J., Lethbridge, T.C. Matwin, S.
  (2003) Applying Data Mining to Software
  Maintenance Records, proc CASCON 2003, Toronto,
  October, IBM, in ACM Digital Library, pp.
  136-148.
• Sayyad Shirabad, J, Lethbridge, T.C., Matwin, S.,
  (2003) Mining the Maintenance History of a
  Legacy Software System, International Conference
  on Software Maintenance (ICSM), Amsterdam, IEEE
  Computer Society, pp. 95-104.
• Jelber Sayyads PhD thesis http//www.site.uottaw
  a.ca/tcl/gradtheses/jsayyad/

10
Project 4 - Proposed work to remodel the UML 2
spec and metamodel in a KB

• Premise
• The UML 2.0 Spec is confusing and inconsistent
• Software engineers could make better use of
  models of they could have a more consistent and
  better specified modelling language and tools
• UML and modelling is a knowledge-rich area, ready
  for knowledge management
• Status
• We are initiating this work with IBM

11
What knowledge is most important to a maintainer

• Simple, navigable facts about their system
• But more than just code knowledge
• Knowledge about
• The domain
• The requirements
• The conceptual framework
• The terminology used
• Simplicity is key

12
How can this knowledge be best captured, and from
where?

• Simple tools
• Extract terminology from documentation and the
  system
• Have software engineers rank and organize it
• They can do 5000 concepts in a day! We did it!

13
How can we make best use of this knowledge?

• In a code exploration tool

14
This presentation will be at

• http//www.site.uottawa.ca/tcl/presentations/kom-
  banff-lethbridge.ppt
• Thank-you!