Sentiment analysis over collaborative relationships in OSS projects

Soure code and dataset.

.
├── README.md
├── process_data.py
├── collaboration.py
├── correlation.py
├── segment.py
├── analysis.py
├── network.py
├── network_plot.py
├── relationships.csv
├── issue-data/
├── Dataset/
    └── grpc/
	    └── grpc_comments.csv
	    └── grpc_polarity.csv
	    └── grpc_label_week.csv
	    └── grpc_label_period.csv
	    └── RQ1/
	    └── RQ2/
      	└── RQ3/
      	└── network/
    └── ipython/
    └── pandas/
    └── openra/
    └── threejs/
├── SentiCR/
    └── classifier.pkl
    └── vector.pkl
    ...
    └── sentiment.py

Data set

The raw data sets crawled from GitHub are in issue-data/.

• mrdoob/three.js
• pandas-dev/pandas
• ipython/ipython
• grpc/grpc
• OpenRA/OpenRA

Project name	URL	Language	Gathered issues	No. of Developers	Star
mrdoob/three.js	https://github.com/mrdoob/three.js	JavaScript	5465	2011	50249
ipython/ipython	https://github.com/scala/scala	Python	10172	3413	13478
pandas-dev/pandas	https://github.com/pandas-dev/pandas	Python	22854	5628	18832
grpc/grpc	https://github.com/grpc/groc	C++	14828	3142	20560
OpenRA/OpenRA	https://github.com/OpenRA/OpenRA	C#	6026	682	6210

Dependencies

• Python >= 3.6
• matplotlib == 3.1.1
• numpy == 1.18.0
• pandas == 0.24.2
• statsmodels == 0.10.2
• scipy == 1.1.0
• networkx == 2.1

Usage

1. Sentiment analysis classifier and vector models are saved in classifier.pkl and vector.pkl. python SentiCR/sentiment.py --train False --prepare False

• prepare will decide whether to preprocess data for sentiment analysis. If true it will generate ($repo)_comments.csv in repo folder from raw issue-data/.

2. Prepare data python process_data.py --repo $repo_name this will group comments by issue, by week and by half-a-year period seperately.

3. Collaboration difference (Q1) python collaboration.py --repo $repo_name python collaboration.py --plot True --path $figure_path this will perform t-test on collaborative relationships vs. non-collaborative relationships, and generate boxplot. if plot is true, this will merge relationships from 5 repos and generate a box plot.

4. Correlation analysis (Q2) python correlation.py --repo $repo_name、--plot True this will perform correlation analysis on 3 factors and generate violin graph and scatter graph.

5. Granger causality test (Q3/1) python segment.py --repo $repo_name python analysis.py --repo $repo_name --lag=2 lag: number of weeks applied to causality test. You can change the number to 2/4/8/12.

6. Network approach (Q3/2) python network.py --repo $repo_name this will generate parameters of three network models python network_plot.py --param average_centrality --path $some_folder