Stack Overflow slides Data Analytics

What StackOverflow Tells Us
About Programming Languages
Rahul Thankachan Nada Aldarrab
Prathmesh Gat
University of Southern California
1

Agenda
1. Introduction
2. Problem Statement
3. Temporal Based Trend Analysis
4. Topic Analysis
5. Predicting Time To Answer
6. Summary and Q&A
3

Introduction
• Dataset Used : Stackoverflow - Internet Archive.
• Why? It is one of the largest developer focused
open collaborative platform currently.
• Through our study we intend to answers some
interesting questions
• Study the rise and fall of popular programming
languages
• Can be used to predict future enhancements
• Study effectiveness of Stack Overflow model
5

Problem Definition
• Basic Analysis: What are the most popular
programming languages?
• What are the trends in programming languages?
• What are the most popular topics discussed in a
programming language?
• Can we accurately predict the time it takes until a
questioner gets an answer?
6

Related Work
Miltiadis Allamanis and Charles Sutton. 2013. Why, When and What: Analyzing Stack Overflow Questions by
Topic, Type & Code In 10th Working Conference on Mining Software Repositories. Mining Challenge. IEEE, pages 53-
56.
• Topic modeling analysis
• Used Latent Dirichlet Allocation (LDA)
• Modeled Java Topics of Questions
• Can evaluate the orthogonality of different languages
• Stack Overflow questions are about the code and are not application domain specific
7

Related Work
V. Bhat, A. Gokhale, R. Jadhav, J. Pudipeddi, and L. Akoglu. Min (e) d your tags: Analysis of question response time in
stackoverflow. In Proceedings of ASONAM 2014, pages 328–335. IEEE, 2014.
• Two linear classifiers: logistic regression and SVM with linear kernel
• Two non- linear classifiers: decision tree (DT) and SVM with radial basis function kernel
8

Related Work
Prediction Accuracy:
9

10
Basic Analysis &
Temporal Trends

StackOverFlow Activity
11
Approx. 45% of all programming languages in world are
discussed on StackOverflow!

Questioners/Answerers Distribution
18

21
Predicting time until
Answer

Approach
Following attributes selected for study:
1. Tag (Only top 10 programming languages)
2. Creation Month
3. Body Length
4. Tag Length
5. Introduced new Nominal class - Time_Answer
6. (less6, bet6and20, 20andmore)
22

Approach
Tools
Weka - Weka is a collection of machine learning
algorithms for data mining tasks.
Data Preprocessing:
Challenging!
23

Approach(Data Pre -processing)
1. Parse all the answers and link first answer’s creation
time to creation time of question. We called this
field delta-answer.
2. Remove all the Questions which had delta answer
negative or zero
3. We developed a Python script which develops .arff
file On the fly (Wish to contribute this file)
24

Evaluation
• Subset Size: 4490947 - Subset - 449000
• Classify response time into 3 types: less than 6
minutes, between 6 and 20 minutes, 20 minutes
and more.
• 10-fold cross-validation
• Results are obtained using different feature
combinations and different classifiers
25

Evaluation
Results of classifier J48 (all Attributes)
26

Evaluation
27
Results of classifier (body_length/ tag_length)

Summary
• We were successfully able to find interesting
temporal trends for major programming languages
• Using tag based topic analysis we were able to find
major discussion topics and to some extent the
difficult topics in a programming language
• Using machine learning techniques we were
successfully able to predict - time to answer with
good accuracy
28

Future Scope of Work
• Contribute the .arff on the fly generator script.
• Adding Parts of speech as an attribute
• Showcasing the results on a website
29

Stack Overflow slides Data Analytics

In this document

More Related Content

What's hot

Viewers also liked

Similar to Stack Overflow slides Data Analytics

Recently uploaded

Stack Overflow slides Data Analytics