This is the repository for our submission to the visualization challenges posed by Mini Challenge 3 of the VAST Challenge 2024, which involves detecting illegal fishing activities within a dynamic network of companies and individuals. The task requires effective anomaly detection in a time-dependent knowledge graph, a scenario where conventional graph visualization tools often fall short due to their limited ability to integrate temporal data and the undefined nature of the anomalies. We demonstrate how to overcome these challenges through well-crafted views in standard software libraries. Our approach involves decomposing the time-dependent knowledge graph into separate time and structure components, as well as providing data-driven guidance for identifying anomalies. These components are then interconnected through extensive interactivity, enabling exploration of anomalies in a complex, temporally evolving network. The source code and a demonstration video are publicly available here.
Our answers for challenge 3 can be found here and our submission video here.
First clone the repository
cd your_parent_folder
git clone https://github.com/MaAllma/Temporal_Knowledge_Graph_Analysis.git
Now download the VAST MC3-data and copy mc3.json and the
Oceanus Information folder to /data (note that the data folder is not tracked by git, so it may need to be
created). (You should now have copied /data/mc3.json and /data/Oceanus Information/*)
Then setup a conda environment for the project:
conda create -n vis_proj python=3.11 tqdm numpy scipy pandas xarray matplotlib seaborn bokeh numba scikit-learn=1.4.2 conda-forge::networkx python-graphblas -c conda -c conda-forge -c pytorch -c nvidia
conda activate vis_proj
and use it as appropiate for your IDE. You may the start the dashboard in this enviornment using:
bokeh serve bokeh_dashboard.py --show
- Information files for the VAST-Challenge are added to the Repository