Hierarchical knowledge graphs (HKGs) are a combined representation of low-level entity relationships and high-level central concepts. We generate these knowledge graphs automatically using a simple parsing algorithm, then extract hierarchies using a dynamic thresholding approach. We evaluate these HKGs using a mixed methods approach. Quantitative data argues that HKGs preserve the transparency advantages of knowledge graphs and structural advantages of hierarchies. Qualitative data triangulates with quantitative observations and provides additional insight into the advantages and disadvantages of both hierarchical and network visualizations.
We find that our hierarchical knowledge graphs preserve many of the previously observed advantages of traditional knowledge graphs, i.e. fewer document views and reduced reading time. Alongside this, hierarchical knowledge graphs introduce an effective hierarchical representation into knowledge graphs.
The main pulications related to HKGs are listed below.
The first paper [1] contrast Knowledge Graphs and Hierarchies as alternative representations of the search results.
The second paper [2] aims at combining the strengths of these two representations into a unified structure, called HKG.
Through a mixed methods evaluation, we showed that HKGs preserve the transparency advantages of knowledge graphs and structural advantages
of hierarchies.
The following questionnaires were used for the in-lab evaluation of HKGs.