Mining Software Data

María Gómez

Software Engineering Course — Summer Semester 2017

 How Software is built is changing…

• Code centric • Data pervasive

• In-lab testing • Debugging in the large

• Centralized development • Distributed development

• Long product cycle • Continuous release

…. ….

Slide adapted from: https://de.slideshare.net/taoxiease/software-mining-and-software-datasets

 Software Data

• Large amount of artefacts are generated in the sw

development process

• Increased amount of data available in software archives

through large open source projects
 Software Decision Making

Sw developers rely on their prior experiences to plan sw

projects, fix bugs, prioritise testing, etc.
Mining Software Repositories (MSR)

Let’s mine software data!

What?

Why?

How?
What is Mining Software Repositories (MSR)?

”The MSR field analyzes rich data available in software repositories

to extract useful and actionable information about software projects
and systems”. (Source: msrconf.org)

DATA Actionable
Software
MINING Information
Data
 What is Mining Software Repositories (MSR)?

Main goals:

• Gather and exploit data produced by developers (and other sw

stakeholders) in the software development process.

• Uses data available in repositories to support development

activities (e.g., defect assignment, software validation, evolution
and planning).

• Discover hidden patterns and trends.

• Transform static record-keeping repositories into active

repositories to guide decision processes.

• Applies data extraction and analysis to make decisions and

predictions.

1 The Road Ahead for Mining Software Repositories. Ahmed E. Hassan.

2 Effective Mining of Software Repositories. Marco D’Ambros, Romain Robbes.
 MSR

• What types of software data are available to mine?

• Which data mining techniques can be used in MSR?

• Which software engineering tasks can be assisted with

MSR?
 MSR

• What types of software data are available to mine?

• Which data mining techniques can be used in MSR?

• Which software engineering tasks can be assisted with

MSR?
 What to mine?

Software repositories refer to artefacts produced and archived

during software development processes by developers and other
stakeholders.
 What to mine?

Different types of repositories1:

Historical Code Runtime

Repositories Repositories Repositories

1 The Road Ahead for Mining Software Repositories. Ahmed E. Hassan.


 What to mine?

Historical Record information about the evolution

Repositories and progress of a project

Examples:
• Version control systems (CVS, SVN, Git, Mercurial)
• Bug repositories (Bugzilla, JIRA)
• Mailing lists (e-mails, wiki pages)
• Development collaboration sites (StackOverflow)
 What to mine?

Code Contain source code of various applications

Repositories Developed by several developers

Examples:
• Code bases (SourceForge, GoogleCode)
• Project ecosystems (GitHub)
 What to mine?

Runtime Contain information about the execution and

Repositories usage of an application

Examples:
• Crash reports
• Field logs
• Execution traces
 What to mine?

Other
Repositories

Examples:
• App Stores (Google Play Store, Apple App Store)
• Contain mobile apps and user feedbacks (reviews, ratings)
 What to mine?

Historical Runtime
Repositories Repositories

Cross-link
of repositories!

Code Other
Repositories Repositories
 Why MSR?

• Better manage software projects

• Produce higher-quality software systems that are delivered on
time and within budget

• Support maintenance of software systems

• Improve software design/reuse

• Learn from past to guide future development

1 MSR Conference: http://2017.msrconf.org/#/home

2 Mining Software Engineering Data. Ahmed E. Hassan & Tao Xie.
 Target Audience
• Software practitioners

• Project Manager

• Developers

• Designers

• Testers

• Usability engineers

• Engineers
 MSR

• What types of software data are available to mine?

• Which software engineering tasks can be assisted

with MSR?

• Which data mining techniques can be used in MSR?

 Applications of MSR
• Estimate developer efforts

• Change impact and propagation

• Risk management (trends)

• Fault analysis and prediction

• Test reduction, minimisation and selection

• Continuous quality assurance

• Post-release maintenance
 Applications of MSR
• New bug report

• Estimate fix effort

• Mark duplicate

• Suggest experts and fix

• New change

• Suggest APIs

• Warn about risky code or bugs

• Suggest locations to co-change

 MSR

• What types of software data are available to mine?

• Which software engineering tasks can be assisted with

MSR?

• Which data mining techniques can be used in MSR?

 MSR Process

Repositories

EXTRACT ANALYZE SYNTHESIZE

Actionable
Information
 MSR Process

Repositories

EXTRACT ANALYZE SYNTHESIZE

Actionable
Information
 Data Extraction

• Extract data from different repositories

• Selection of input data

• Processing (e.g., filtering)

• Constraints to help with scalability

 MSR Process

Repositories

EXTRACT ANALYZE SYNTHESIZE

Actionable
Information
 Data Analysis

• Process the data

• Link data between repositories

• Empirical analysis to the data

Types of Empirical Analysis

Different types of empirical analysis can be performed in

repositories:

• Quantitative vs qualitative

• Regression models

• Grounded theory

• Machine learning/data mining

Types of Empirical Analysis
Quantitative vs qualitative
 Types of Empirical Analysis
Quantitative vs qualitative

Quantitative Qualitative

Data is numerical Data non-numerical

Data can be measured Data can be observed
 Types of Empirical Analysis
Quantitative vs qualitative

Example quantitative study:

Do performance bugs take more time to fix?
Are performance bugs fixed by more experienced developers?

Example qualitative study:

What are the advantages/disadvantages of shared code
ownership from the developers perspective?
Types of Empirical Analysis
Regression models
• Estimate relationship among variables
• Widely used for prediction and forecasting

Example:
What factors contribute to delays on bug fixing time most?
Types of Empirical Analysis

Grounded theory

• Building theory from data

• Discovery of emerging patterns in data
 Types of Empirical Analysis
Grounded theory

Figure source: https://www.researchgate.net/figure/222301824_fig1_Fig-1-Basic-process-of-the-Grounded-Theory-approach

Types of Empirical Analysis

Machine learning/data mining techniques

• Association Rules and Frequent Patterns

• Classification

• Clustering
 Data mining techniques
Association Rules and Frequent Patterns
• Find frequent patterns in a database
• Itemset: set of items
• Support of itemsets
• Confidence of rules

Image source: https://image.slidesharecdn.com/3-150328084211-conversion-gate01/95/31-mining-frequent-patterns-with-association-rulesmca4-4-638.jpg?cb=1427532681

 Data mining techniques
Classification

• Supervised learning

1. Construct model with labeled objects (training set).

2. Apply model to unlabelled objects.

 Data mining techniques
Clustering
• Unsupervised learning (no predefined classes)

• Group similar data

 Analysis Tools

Data mining and analysis tools:

• R
http://www.r-project.org/
Free software for statistical computing and graphics

• Weka
http://www.cs.waikato.ac.nz/ml/weka/
Open-source tool containing a collection of machine learning and
data mining algorithms.
 MSR Process

Repositories

EXTRACT ANALYZE SYNTHESIZE

Actionable
Information
 Data Synthesis

• Report / visualisation of outcome

• Understand the needs of practitioners

• Help practitioners to make decisions

• Don’t replace them!
 Actionable Outputs

• Developer feedback

• Bug prediction

• Quality assurance

• Architecture analysis

• ………
What can we learn from
software data?

MSR Application Examples

 Can we predict bugs?

• Link bug fixes to source code changes

• Eclipse/Mozilla repos and bug-trackers
• Correlations found!

When do changes induce fixes? Jacek Sliwerski, Thomas Zimmermann and Andreas Zeller. (MSR’ 05)
 Can we predict bugs? (2)

Example source: https://de.slideshare.net/taoxiease/software-mining-and-software-datasets

 How Long will it Take to Fix this Bug?

• Predicting effort to fix a bug

• Mine bug databases
• Text similarity to identify reports closely related

How Long will it Take to Fix This Bug? C. WeiB, R. Premraj, T. Zimmermann, A. Zeller. (MSR’ 07)
 Can we identify duplicate bug reports?

• Mine bug repositories (e.g., Bugzilla, Jira)

• Use information retrieval to find similar reports and rank them.

Search-Based Duplicate Defect Detection: An Industrial Experience. Amoui, M., Kaushik, N., Al-Dabbagh, A., Tahvildari, L., Li, S., & Liu, W. (MSR’13)
 Change Propagation
How does a change in one source code entity propagate to other entities?

• Predict change propagation

• Mine association rules from change history

Predicting Change Propagation in Software Systems. Ahmed E. Hassan and Richard C. Holt (ICSM ’04)
 Classify Changes as Buggy or Clean
• Can we warn developers that there is a bug in a change’’?

• Identifying bug-introducing changes from bug-fix data

Automatic Identification of Bug-Introducing Changes. Kim, S., Zimmermann, T., Pan, K., & James Jr, E. (ASE’ 06)
 Classify Changes as Buggy or Clean

Automatic Identification of Bug-Introducing Changes. Kim, S., Zimmermann, T., Pan, K., & James Jr, E. (ASE’ 06)
 Classification of security bug reports

Example source: https://de.slideshare.net/taoxiease/software-mining-and-software-datasets

Mining questions about software energy consumption

• Mine communities (StackOverflow)

• Use thematic analysis (e.g. LDA, Classifier) to find common themes in

questions & answers

• Interpret themes

Mining questions about software energy consumption. Pinto, G., Castor, F., & Liu, Y. D. (MSR’ 14)
 API change and fault proneness
impact success
• Relationship between success of Android apps and Android API
instability

• Measure success through user ratings in app store

• Measure fault-proneness through number of bugs fixed in the used

APIs

API change and fault proneness: a threat to the success of Android apps. M. Linares et al. (FSE’13)
 Recommending and Localizing Change
Requests for Mobile Apps based on
User Reviews
• Automatic classification of user reviews from Google Play store

• Link to the source code entities to be changed

• Recommend developers changes to sw artefacts

Recommending and Localizing Change Requests for Mobile Apps based on User Reviews. F. Palomba et. al. (ICSE’17)
 MSR in Practice

Slide extracted from: https://de.slideshare.net/taoxiease/software-mining-and-software-datasets

 Tools for Mining Software
Repositories

• Available mining tools

• Libresoft Tools. http://tools.libresoft.es/

• CVSAnaly. VS/SVN/Git repository log parser

• MLStats. Mailman and Mboxes parser

• Bicho. Bugzilla and SF.net tracker parser

 MSR Repositories

Data Repositories available online:

• FLOSSmole repository of open source snapshots. flossmole.org/

• Github. http://www.ghtorrent.org
• iBUGS. www.st.cs.uni-saarland.de/ibugs/
• MetricsGrimoire toolset. https://metricsgrimoire.github.io
• PROMISE repository. http://openscience.us/repo/
• Software-artifact Infrastructure Repository. http://sir.unl.edu/portal/index.php
• Ultimate Debian Database. https://wiki.debian.org/UltimateDebianDatabase
• Apache SVN commits. https://github.com/monperrus/apache-svn-commits
• Socorro: Mozilla Crash Stats. https://wiki.mozilla.org/Socorro
 References
• The International Conference on Mining Software Repositories.
2017.msrconf.org

• Mining Software Engineering Data. Ahmed E. Hassan & Tao Xie.

• The Road Ahead for Mining Software Repositories. Ahmed E.

Hassan

• Software Intelligence: The Future of Mining Software Engineering

Data. Ahmed E. Hassan & Tao Xie.

• Effective Mining of Software Repositories. M. D’Ambros & Romain

Robbes.