On this research project page we present the results of our research about Human Associations in the Semantic Web.
Associations are one of the building blocks of human intelligence, thinking, context forming and everyday communication. They have been the focus of psychological research for a long time and there are excellent datasets such as the Edinburgh Associative Thesaurus (EAT).
The Semantic Web is a vision of a World Wide Web for interlinked data, not only web pages for humans. Following this vision, the Linked Data movement has brought us many cool interlinked datasets. A prominent example of such a dataset is DBpedia, which is extracted from Wikipedia.
We also extracted graph patterns for these semantic associations from DBpedia. We developed and applied an evolutionary graph pattern learner which is able to learn SPARQL patterns for a given list of source-target-pairs. You can find a visualisation of its results below.
The generated association vocabulary ( RDF | OWL | TTL ) has its own doc page.
It mainly consists of the classes Term, for a stimulus or response, and Association which connects them to their count and frequency. It also defines classes Mapping and VerifiedMapping to map associations from one dataset to another with the mapped to property.
An example for all this can be found in the following drawing. In the middle you can see the stimulus "pupil", which can lead to the responses "eye" (left) and "school" (right). Both associations are mapped to a corresponding association node which also contains information about ther counts and frequencies. The one on the left further was mapped to a semantic association between the two DBpedia entities dbr:Pupil and dbr:Eye. As you can see in this process the ambiguous term "pupil" was disambiguated to the right semantic entity dbr:Pupil.
See our papers in the publications section for more details.
In order to find graph patterns for associations in DBpedia, we developed an evolutionary graph pattern learner (see paper). It takes an input list of source-target-pairs (such as the stimulus-response-pairs of the semantic associations above) and learns SPARQL queries for them from a given SPARQL endpoint.
The source-code for our algorithm can be found on GitHub.
We ran our pattern learner against a local Linked Data endpoint filled with 7.9 G triples from the heart of the LOD-Cloud.
The resulting learned graph patterns are available in an interactive visualisation:
In the context of our work the following papers have been published:
J. Hees, R. Bauer, J. Folz, D. Borth & A. Dengel -
Edinburgh Associative Thesaurus as RDF and DBpedia Mapping.
Proceedings of ESWC 2016 SE, Herarklion, Crete.
LNCS, Springer.
( Paper | ESWC | Springer | Poster | Extended Paper )
J. Hees, R. Bauer, J. Folz, D. Borth & A. Dengel
An Evolutionary Algorithm to Learn SPARQL Queries for Source-Target-Pairs - Finding Patterns for Human Associations in DBpedia.
Proceedings of EKAW 2016, Bologna, Italy.
LNCS, Springer.
Have a question, some idea, feedback or want to collaborate?
Just drop us a line, we don't bite ;)