Detecting Anomalies in Graph Networks on Digital Markets
Author Information
Author(s): Agata Skorupka
Primary Institution: SGH Warsaw School of Economics, Warsaw, Poland
Hypothesis
The study aims to find the most efficient method for detecting anomalous activities in digital markets using graph-based methods.
Conclusion
Graph-based methods outperform traditional text-based methods for detecting anomalies in digital markets.
Supporting Evidence
- Graph-based methods showed better predictive performance than text data.
- Node statistics algorithms reduced computation time significantly.
- Models based on node statistics consistently ranked high across datasets.
Takeaway
This study looks for ways to find bad users in online markets, like bots or fraudsters, using connections between users instead of just their messages.
Methodology
The study utilized three publicly accessible graph-based datasets and compared various graph-based methods for anomaly detection.
Potential Biases
The manual annotation process for labeling bots may introduce bias.
Limitations
The study is limited to supervised problems and does not address unsupervised learning.
Participant Demographics
The study analyzed Twitter users and Bitcoin market participants.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website