SwedenTechnology and telecom2011 – ongoingEricsson
TRR machine-learning assignment of software bug reports
Ericsson's TRR automatically assigned 30% of incoming bug reports with 75% accuracy
Problem
Experienced Ericsson engineers spent substantial time assigning incoming bug reports to development teams. Their specialist knowledge became a bottleneck.
Approach
Research began in 2011 and developed into Ericsson's internal TRR product. A logistic-regression classifier used report descriptions and structured attributes. Automatic routing began in April 2019 and was restricted to sufficiently confident predictions.
Results
Takeaway
Selective automatic routing left other reports for human judgment. Teams working on lower-level modules experienced drawbacks that were less apparent in higher-level modules.
Categories
- Size
- Large enterprise
- Industry
- Technology and telecom
- Business function
- IT and R&D
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Partner co-development
- Evidence
- Company's own claim
Sources
Source summaries
The researchers combined bug-tracking data with interviews to examine TRR's development and use. They reported automatic assignment of 30% of reports at 75% accuracy and around 21% faster resolution for automatically routed reports. Ericsson employees coauthored the study.
Full details are in the original sources.
