Evaluation of alignment methods to support the assessment of similarity between e-commerce knowledge graphs

Enterprise knowledge graphs combine data from multiple sources, and the structure and granularity of the resulting graphs can vary widely. The data in knowledge graphs is typically structured by schemas, mainly ontologies. Finding the alignments between content from different graphs allows to reduce redundancy, detect inconsistencies, and improve the use of the data. Recently, many graph alignment methods have been proposed, usually tested on general knowledge bases, so there is very little knowledge about the effectiveness of the methods in specific domains. Moreover, we are also interested in studying how these methods capture the semantic similarity between the schemas of the knowledge graphs. In this article, we study both aspects using 25 graph alignment methods on datasets related to e-commerce activities of organizations, such as product sales or customer satisfaction. We have developed a pipeline for generating and evaluating the results. The results show that the alignment methods can be used as a semantic similarity system between the ontology-dataset pairs that generate the knowledge graphs, that AttrE and BootEA are the most effective and robust methods across the different datasets, that the complexity of the structure of the ontology has a clear impact on the effectiveness of the methods, and that the results of the alignment experiments generate information about the similarity of the ontologies and reveal which parts of them should be better modeled to increase the performance of the methods. • Methodological framework for systematic analysis of entity alignment experiments. • Evaluation of 25 methods for supporting semantic similarity analysis. • AttrE and BootEA are the best performing methods based on standardized metrics. • The structure of the ontology influences the results of the graph alignment methods. • Not all class attributes or relations have equal influence on the alignment results.