SE Data Mining Based Software Defection Detection

1
SE Data Mining Based Software Defection Detection

• ???

2
Outline

• Significance
• Background
• Statistic of Publications
• Typical Method
• Future Work

3
Significance

• More complex software
• More Critical Application
• Faster bug generation
• (compare with slow manual detection)
• Abundant SE data

4
Eg Bugzilla_at_Mozilla

• About 100 defections reported per day
• Few fixed
• More than 20,000 Bugs left

5
Background

• SE Data
• Documentation
• Codes
• Comments
• Debug Info.
• Build Info.
• Communication Info.

6
Background

• Software Defections
• Code Defection
• Interface Defection
• Check Defection
• Build Defection
• Document Defection

7
Sum of papers
 
8
Sum of papers
9
CVS Search

• Assumption developers are more likely to comment
  changes.
• Track changes and define fragments
• Combine CVS and comments
• Goal query comments help developers to
  understand codes

10
Bug Classify

• Assumption too many reports to classify manually
• Machine learning for labeling reports
• (Sth. Nature language mining)
• Goal help to classify bugs and decide who shall
  fix it
• Evaluation accuracy

11
Mining Mail Box

• Assumption the structure of developer team
  reflect the structure of software
• Study the graph structure of Mozilla mailbox

12
Bugs or bad comments?

• Assumption developers are lazier to maintain
  comments than codes.
• Finding and merging topics
• Classify comments by topics
• Discover comment rules via defined code rules
• Goal find inconsistencies
• Evaluation number accuracy

13
Partial Order of API Concepts

• Assumption most codes are well written
• Why partial order?
• What is support?

14
Partial Order of API Details

• Static path
• Growing
• Edge merging
• Triggers
• PC chains
• Deep first search
• Verify sample
• Violates are bugs

15
Partial Order of APIEvaluation

• Find more traces
• Compress time consumption
• Avoid false traces
• Find more bugs

16
Future Work

• Better means of classification
• BBS IM message mining
• Documents mining to find interface defection
• Mining software leaks to find check defection
• How to automatically classify bugs

17
Thanks!