• Introduction
• Demonstration
• Version History
• Dependencies
• Setup
• Application Start/Stop
• Source Download

This repository contains the source code of a no-code platform for tie prediction analysis in social media networks. This platform is open source and uses the well-established python packages "NetworkX" (https://networkx.org/) and "Scikit-Learn" as a foundation (https://scikit-learn.org/stable/). Contributions to this public repository are welcome, as
they may enhance this tool.

The platform enables conducting tie prediction analysis without the need to resort to coding. Common tie prediction approaches can be configured via the platform's dashboard. The open source paradigm enables augmenting the platform with further prediciton approaches or adjusting already incorporated approaches. This is facilitated by the fact that this tool relies on well-established and well-documented python packages (see above).

The following steps illustrate an exemplary no-code tie prediction analysis in social media networks.

• Navigate to the tie prediction dashboard and select "+" to open the following window:

• Import a network dataset in a suitable format (see the folder "2-sampledata" for exemplary datasets)
• Please Note: Ensure the node ids of your network dataset start with index 1

• Select the created project, which leads to the following window:

• Select "Settings"

• Add topology-based tie prediction methods via "+" (e.g., Adamic-Adar)

• Add social theory-based tie prediction methods via "+" (e.g., homophily)

• Add a machine learning classifier via "Create" (e.g., decision tree)

• Configure further attributes, if possible (e.g., homophily weightings)

• Configure the evaluation setup (e.g., the train-test-split ratio)

• Save the prediction setup with the corresponding button

• Execute the prediction with the corresponding button "Predict", which should lead to the following window:

• This window displays the achieved AUC and whether machine learning was used to derive this AUC value
• Moreover, it displays when this AUC value was achieved
• Lastly, it provides further data about the considered social media network

• Navigate to the visualization tab to see the prediction results within the imported network:

• Specific dyads can be analyzed (click on the first node, hold CTRL, then click on the second node).

• Green: True positive predictions. Red: False positive/false negative predictions. Blue: True negative predictions.

• Further displays which approach(es) informed the prediction results. For instance, in the figure above, the decision tree approach was able to predict one of the two directed ties in the observed dyad, leading to an accuracy of 0,5.

• Can be deactivated to substantially increase prediction velocity (see the folder linkprediction\prediction_methods)

• Navigate to the evaluation tab to see the corresponding achieved AUC scores:

• Displays the achieved AUC scores for each approach both as a plot and as a table.
• If a train-test-split was chosen, the buttons on the right corner can be used to view the train/test AUC scores.

• Current version: version-1.0
• Version history: (this is the first productive version)
• For planned features in future versions, see the document "1)roadmap.txt" in the folder "3-documentation"

• node.js (see https://nodejs.org/ or the folder "4-resources")
• python (see https://www.python.org/ or the folder "4-resources")
• postreSQL (see https://www.postgresql.org/)
• pgadmin (optional, if a database interface is desired) (see https://www.pgadmin.org/)
• anaconda (optional, if source code adjustments are planned) (see https://www.anaconda.com/)

• Download and install postgresSQL (see https://www.postgresql.org/)

  • Keep the default installation settings
  • Add postgresSQL bin file location (e.g., C:\Program Files\PostgreSQL\13\bin) as PATH variable

• Install node.js (see https://nodejs.org/ or the folder "4-resources")

  • Keep the default installation settings

• You can setup the application automatically by running the application-install.bat file (64 bit or 32 bit)
  • Please note: You can find the application-start.bat file in the folder "5-setup-start"
  • Please note: The postgresSQL server should start on windows startup (see under services.msc)
  • Please note: The password is 12345

• Open cmd in the folder "1-database" (by typing "cmd" in the file explorer bar)
• Run: "psql.exe -h 127.0.0.1 -p 5432 -U postgres -d postgres -c "CREATE EXTENSION IF NOT EXISTS ""uuid-ossp"";"
• Run twice: "psql.exe -h 127.0.0.1 -p 5432 -U postgres -d postgres -f db-schema.sql"
  • Please Note: The password is 12345

• Open cmd or terminal in the root repository directory
• Navigate to the directory "linkprediction/frontend/angular"
• Install the node packages with "npm install"

• Install python (see https://www.python.org/ or the folder "4-resources")
• Open cmd or terminal the "4-resources" repository directory
• Install the python libraries with "pip install -r requirements.txt"

• You can start the components automatically by running the -start.bat files in the "5-setup-start" folder
• Open the frontend-start.bat and the backend-start.bat files and press "Enter" twice in each cmd
• Open the frontend-dashboard.url file
• You can stop the components by pressing "CRTL+C" twice in the corresponding cmd windows
• The dashboard is accessible via the URL "localhost:4200"
• The backend is accessible via the URL "localhost:8080/api"

• Open cmd or terminal in the root repository directory
• Navigate to the directory "linkprediction/frontend/angular"
• Start the frontend in local server with"npm run ng serve"
• Open the frontend-dashboard.url file
• The dashboard is accessible via the URL "localhost:4200"
• You can stop the componenent by closing the cmd window

• Open command line tool in the cloned repository directory
• Start backend in local server with "python -m linkprediction"
• The backend is accessible via the URL "localhost:8080/api"
• You can stop the componenent by closing the cmd window