A hybrid AI system capable of identifying bias patterns in datasets used by predictive AI models. The proposed system captures knowledge about dataset characteristics and represents it as factual statements in a knowledge graph.

The proposed hybrid AI system is used to trace the PSL implementation of the AI model for fake news detection (https://doi.org/10.1145/3340531.3412066).

This repository contains following contents:

1. DataSources: The data files generated to trace the fake news model and dataset defined in: https://github.com/linqs/chowdhury-cikm20.
2. Doc-BiasKG: schema, mapping rules, resulting experiments.
3. Folder sparql_queries contains all the necessary bias analysis performed over the generated Doc-Bias KG.
4. DocBias_over_InterpretME: Doc-Bias analysis applied over a principled approach, InterpretME. The necessary instructions for execution of InterpretME pipeline and running sparql queries over the InterpretME KG is available in this folder.

Clone the repository

git clone [email protected]:SDM-TIB/DocBiasKG.git

Install necessary dependencies for Doc-Bias approach, create a virtual environment and run the following command:

pip install -r requirements.txt

References

[1] Marco Ribeiro, Sameer Singh, and Carlos Guestrin. "Why Should I Trust You?": Explaining the Predictions of Any Classifier. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD '16). ACM. 2016. DOI: 10.1145/2939672.2939778

[2] Yashrajsinh Chudasama, Disha Purohit, Philipp D. Rohde, Julian Gercke, Maria-Esther Vidal. "InterpretME: A Tool for Interpretations of Machine Learning Models Over Knowledge Graphs" In Semantic Web Journal, Special Issue on Tools & Systems. DOI: 10.3233/SW-233511