Journal first presentation at ICSE'17 in Buenos Aires, Argentina.
M. Borg, K. Wnuk, B. Regnell, and P. Runeson. Supporting Change Impact Analysis Using a Recommendation System: An Industrial Case Study in a Safety-Critical Context, IEEE Transactions on Software Engineering, 43(6), pp. 675-700, 2017.
Supporting Change Impact Analysis Using a Recommendation System - An Industrial Case Study in a Safety-Critical Context
1. SUPPORTING CHANGE
IMPACT ANALYSIS USING
A RECOMMENDATION
SYSTEM
- An Industrial Case Study in a Safety-
Critical Context
@mrksbrg
Markus Borg, Krzysztof Wnuk, Björn Regnell,
Per Runeson
Research Institutes of Sweden
Swedish Institute of Computer Science
4. Development engineer, ABB, Malmö, Sweden
Editor and compiler development
Safety-critical systems
PhD student, Lund University, Sweden
Machine learning for software engineering
Bug reports and traceability
Senior researcher, RISE SICS AB, Lund, Sweden
Software engineering for machine learning
V&V for self-driving cars
10. System evolution since
the 1980s
~ 2 MLOC C/C++
Safety certification
(SIL-2)
Evolving automation system
Formal development
process
V model, stage-gate
process
~200 SW engineers
Main sites in Sweden
and India
13. Automated Change Impact Analysis
Intuitive tool to jump start analyses based on
historical data
Goal: faster & more accurate analyses
Approach:
1. Mine the history
2. Recommend impact
14. Part 1: Construct network of
previously reported impact
Index textual data with
16. Find similar bugs using Apache Lucene
Part 2: Recommend impact
Design Doc. X.Y
Req. X.Y
Test case UTC56
Req. Z.Y
Design Doc. X.Y
Follow links to identify candidate impact set
17. Use centrality measures to rank candidate impact
1. Requirement X.Y
2. Design Document X.Y
3. Test case UTC56
4. Design Document X.Y
5. Requirement Z.Y
Find similar bugs using Apache Lucene
Part 2: Recommend impact
Follow links to identify candidate impact set
23. Industrial case study - Utility
Selected two teams
Conducted interviews
Deployed ImpRec
Collected data for 3-9 months
Logged every click
Calculated correctness metrics
Conducted interviews