Relation Hierarchy

Introduction

Relation hierarchy is a large collection of relations (properties) collected from multiple knowledge bases like Wikipedia Infobox, Wikidata, and DBpedia. It contains individual relation hierarchy from each of the knowledge base and also joint hierarchy for combination of knowledge bases. The combined hierarchy of all 3 resources contains 623 relations.

Check our CODS-COMAD for complete details.

Example

Relation Hierarchy is created by finding taxonomical (is-a) relation between the relations.
Each path from root node to leaf node has following nodes:

• rel
• source_entity_type
• source_entity_type - target_entity_type
• [... ancestor_relations_node(s)] (if any)
• relation

Hierarchy Visualisation

Visualisation of all the relation hierarchies is available at this portal.

Note: Certain Browsers may fail to display the hierarchical Tree due to CORS policy.In that case go to this link,select "Request temporary access to demo server" and try again.

Statistics

Table below shows basic statistics of three individual hierarchies Hi (Wikipedia Infobox), Hd (DBpedia), Hw (Wikidata), and the common hierarchy H.

Figure (left) below shows the distribution of relations from different resources in each of the 9 buckets. And Figure (right) shows the number of relation in each bucket in the combined relation hierarchy.

Usage

Relation hierarchy was created with aim to advance the reseach on hierarchical relation learning. We are using this hierarchy to create new resource for relation learning. To get started you can go through our work on Relation-Hierarchy.

Download all list the relations, and their mapping with KB relations. [relation-hierarchy.zip]

Download all DBpedia tuples aligned with relation hierarchy. [DBpedia-tuples.zip] [Size: compressed(64 MB), uncompressed(353 MB)] [Number of Tuples: 5.5 Millions]

Download all Wikidata tuples aligned with relation hierarchy. [wikidata-tuples.zip] [Size: compressed(166 MB), uncompressed(1 GB)] [Number of Tuples: 15 Millions]

Citations

Please cite the following paper if you use this website for research