Centrality measures and prediction tasks for signed networks

• Nejat Arinik nejat.arinik@inrae.fr
• Vincent Labatut vincent.labatut@univ-avignon.fr
• Rosa Figueiredo rosa.figueiredo@univ-avignon.fr

This set of Python scripts/modules is designed for two main purposes:

• implementing centrality measures for signed networks
• performing some prediction tasks, such as classification and regression, based on the features extracted from signed centrality measures or other graph-related measures/statistics.

In order to compute centrality measures or to perform some prediction tasks, we include a sample dataset in the folder in. It is a part of the full training dataset used in our article: Space of optimal solutions of the Correlation Clustering problem (Arinik, Nejat; Labatut, Vincent, Figueiredo, Rosa (2020): Space of optimal solutions of the Correlation Clustering problem on Complete Signed Graphs. figshare. We also extended this dataset with incomplete signed networks. The final dataset can be found on FigShare.

• Folder in: input signed networks.

• Folder out: contains the files produced by our scripts

• Folder src:

  • Folder descriptors: this folder contains three groups of measures: 1) centrality, 2) node embeddings, and 3) graph embeddings.
  • Folder collect: the scripts of this folder aim at collecting the statistics/measures which are already computed and stored in folders.
  • Folder prediction: this folder contains the scripts performing the prediction tasks, such as classification and regression.
  • Folder stats: this folder contains some graph-related statistics, e.g. spectral, structural, structural balance.

• Folder tests:

  • Folder centrality: All the unit and integration tests related to signed centrality measures.
  • Folder stats: All the unit and integration tests related to graph-related statistics.
  • Folder graph_embeddings: All the unit and integration tests related to signed graph embeddings.
  • Folder node_embeddings: All the unit and integration tests related to signed node embeddings.

• There are 4 main files.

  • main_feature_and_output_extraction.py: It processes the features and the outputs.
  • main_binary_classification.py: It performs the binary classification task.
  • main_ordinal_classification.py: It performs the ordinal classification task.
  • main_regression.py: It performs the regression task.

• Install Python (tested with Python 3.8.12)

• Install dependencies using the following command:

  pip install -r requirements.txt
  

• Download this project from GitHub

• You need to retrieve the data from Figshare. Download and untar Input Signed Networks.tar.gz and Evaluation Results.tar.gz. Place them into the in and output/evaluate-partitions folders, respectively. Finally, we might need to configure the input parameters, i.e. the global variables such as GRAPH_SIZES, in the main files. You do not need to change these input parameters, if you work with the whole dataset.

• Go to the folder src.

• We run the file main_feature_and_output_extraction.py inside the folder src. Or, you can configure the PYTHONPATH variable, if you do not want to run it from the folder src. Note that it can take several hours.

• Apply the use case 2

• If you skip the use case 1, then you need to download and untar csv.tar.gz from Figshare. Place it into the out/csv folder.

• Go to the folder src.

• We run the files main_binary_classification.py, main_ordinal_classification.py and main_regression.py inside the folder src. Or, you can configure the PYTHONPATH variable, if you do not want to run it from the folder src.

• add more signed centrality measures
• add more graph-related statistics
• add some signed node embeddings
• add some signed graph embeddings
• add ordinal logistic regression from the package statsmodels
• add recursive feature selection method from the package scikit-learn