Finding Explanations of Entity Relatedness in Graphs: A Survey

TL;DR

This survey reviews methods for explaining entity relatedness in graphs, focusing on path-based and node relevance-based approaches, highlighting their differences, applications, and key concepts involved.

Contribution

It provides a comprehensive overview and classification of existing solutions for entity relatedness explanations in graphs, aiding understanding and future research.

Findings

Path-based explanations rank paths by importance and informativeness.

Node relevance measures help identify subgraphs that explain relatedness.

The survey contrasts different approaches and discusses key concepts.

Abstract

Analysing and explaining relationships between entities in a graph is a fundamental problem associated with many practical applications. For example, a graph of biological pathways can be used for discovering a previously unknown relationship between two proteins. Domain experts, however, may be reluctant to trust such a discovery without a detailed explanation as to why exactly the two proteins are deemed related in the graph. This paper provides an overview of the types of solutions, their associated methods and strategies, that have been proposed for finding entity relatedness explanations in graphs. The first type of solution relies on information inherent to the paths connecting the entities. This type of solution provides entity relatedness explanations in the form of a list of ranked paths. The rank of a path is measured in terms of importance, uniqueness, novelty and informativeness. The second type of solution relies on measures of node relevance. In this case, the relevance of nodes is measured w.r.t. the entities of interest, and relatedness explanations are provided in the form of a subgraph that maximises node relevance scores. This paper uses this classification of approaches to discuss and contrast some of the key concepts that guide different solutions to the problem of entity relatedness explanation in graphs.