Counter Effect Rules Mining in Knowledge Graphs

Discovering causal relationships is the goal of many experiments in science. Such a relationship indicates that a variation in an attribute, i.e., the treatment, implies a variation, i.e., has an effect, on another attribute, i.e., the outcome. Mining causal relationships have been studied in a recent approach in Knowledge Graphs, where differential causal rules are mined. Such rules express an effect of a treatment on a subset of instances described by a graph pattern named strata. However, these rules can be difficult to interpret, especially when a treatment has different effects depending on the strata it is expressed on. This paper presents counter effect rules that can be discovered from differential causal rules to facilitate their interpretation. This representation allows to point out the strata that lead to opposite effects for the same treatment. Our experiment shows that counter effect rules can be discovered on a real dataset.