Method of and system for ranking subgraphs as potential explanations for graph classification

A system for ranking subgraphs as potential explanations for a labelled edge type class, the system comprising a processor, the processor having access to a machine learning model having been trained to embed graphs,

• the processor being operatively connected to a non-transitory storage medium comprising instructions, the processor upon executing the instructions, being configured for:
• obtaining a first graph, the first graph being a first type of representation of a set of labelled digital items, each labelled digital item being represented as a respective entity node connected via at least one respective labelled edge type to a respective value node;
• obtaining a second graph, the second graph being a second type representation of relations in the set of labelled digital items, the second graph comprising at least one of:
• a first edge type between a first entity node and a second entity node,
• a second edge type between a third entity node and a first value node, and
• a third edge type between a second value node and a third value node;
• generating a combined graph by merging the first graph and the second graph based on the at least one of the first edge type, the second edge type and the third edge type;
• obtaining a set of unlabelled digital items, each unlabelled digital item comprising at least one value;
• matching at least one unlabelled digital item to a respective subgraph of the combined graph to obtain a set of matched subgraphs, the respective subgraph being representative of at least a portion of the at least one unlabelled digital item;
• embedding, using the machine learning model, the combined graph to obtain a set of vectors encoding the combined graph in a multidimensional space;
• determining a respective expressivity score between a given vector corresponding to a respective labelled edge type in the combined graph, and a respective average vector corresponding to a respective matched subgraph of the set of matched subgraphs in the combined graph to obtain, for each respective labelled edge type, a respective subset of expressivity scores associated with respective matched subgraphs;
• and
• ranking, based on the respective subset of expressivity scores, each respective matched subgraph to obtain a ranked set of potential explanation subgraphs for the respective labelled edge type.