A Business Intelligence Tool for Explaining Similarity

Abstract

Agile Business often requires to identify similar objects (firms, providers, end users, products) between an older business domain and a newer one. Data-driven tools for aggregating similar resources are nowadays often used in Business Intelligence applications, and a large majority of them involve Machine Learning techniques based on similarity metrics. However effective, the mathematics such tools are based on does not lend itself to human-readable explanations of their results, leaving a manager using them in a “take it as is”-or-not dilemma. To increase trust in such tools, we propose and implement a general method to explain the similarity of a given group of RDF resources. Our tool is based on the theory of Least Common Subsumers (LCS), and can be applied to every domain requiring the comparison of RDF resources, including business organizations. Given a set of RDF resources found to be similar by Data-driven tools, we first compute the LCS of the resources, which is a generic RDF resource describing the features shared by the group recursively—i.e., at any depth in feature paths. Subsequently, we translate the LCS in English common language. Being agnostic to the aggregation criteria, our implementation can be pipelined with every other aggregation tool. To prove this, we cascade an implementation of our method to (i) the comparison of contracting processes in Public Procurement (using TheyBuyForYou), and (ii) the comparison and clustering of drugs (using k-Means) in Drugbank. For both applications, we present a fairly readable description of the commonalities of the cluster given as input.