Automatic Facet Generation and Selection over Knowledge Graphs

Abstract

Recently, The Linked Data Cloud is constantly growing, because it has become a valuable source of semantic data enrichment. Therefore, methods to efficiently and easily explore semantic data are required. Faceted Browsing is a widely used approach to partition resource sets based on various aspects. It allows exploratory search over unknown data schemata without exhaustive domain knowledge. Generally, users are presented with an overview of resources' attributes, called facets, that can be leveraged to narrow down the description of their information need. However, though manual facet predefinition is a valid approach, it becomes impractical, since semantic knowledge graphs (KGs) are often continuously changing. Furthermore, large-scale and heterogeneous KGs induce a large number of possible facets. We propose an approach for automatic facet generation, ranking and selection over KGs. Our ranking is based on metrics that both rate facets individually as well as collectively. We also exploit indirect properties to find better categorizations and thus improve facets' usefulness. Finally, we integrate all proposed methods in an overall automatic facet generation workflow.