Causal Discovery in Knowledge Graphs by Exploiting Asymmetric Properties of Non-Gaussian Distributions

TL;DR

This paper introduces a novel hybrid method for discovering cause-effect relationships in knowledge graphs by leveraging non-Gaussian models and tensor decomposition techniques, addressing the lack of domain knowledge in causal inference.

Contribution

It presents a new framework combining tensor decomposition and non-Gaussian causal discovery algorithms to infer causal structures in knowledge graphs.

Findings

Effective identification of causal orderings in knowledge graphs

Integration of tensor decomposition with causal discovery algorithms

Potential for improved causal inference without domain knowledge

Abstract

In recent years, causal modelling has been used widely to improve generalization and to provide interpretability in machine learning models. To determine cause-effect relationships in the absence of a randomized trial, we can model causal systems with counterfactuals and interventions given enough domain knowledge. However, there are several cases where domain knowledge is almost absent and the only recourse is using a statistical method to estimate causal relationships. While there have been several works done in estimating causal relationships in unstructured data, we are yet to find a well-defined framework for estimating causal relationships in Knowledge Graphs (KG). It is commonly used to provide a semantic framework for data with complex inter-domain relationships. In this work, we define a hybrid approach that allows us to discover cause-effect relationships in KG. The proposed approach is based around the finding of the instantaneous causal structure of a non-experimental matrix using a non-Gaussian model, i.e; finding the causal ordering of the variables in a non-Gaussian setting. The non-experimental matrix is a low-dimensional tensor projection obtained by decomposing the adjacency tensor of a KG. We use two different pre-existing algorithms, one for the causal discovery and the other for decomposing the KG and combining them to get the causal structure in a KG.