John Kausch deposited Testing Ontology Embedding Visualization in the group CityLIS on Humanities Commons 3 months ago
This dissertation presents an experiment conducted with human participants on human-information interaction with visualizations of ontologies. The research question is whether embedding visualizations or graph based visualizations lead to better task performance for human-information interaction. A literature review of word embeddings, information retrieval applications, cartesian and radial visualizations, and knowledge graph visualizations is conducted. This literature review is grounded in a facet analysis of the intersecting topics of the central research question. The context of embeddings as used for information retrieval in the 20th century, as opposed to more recent 21st century inventions such as Google’s word2vec is explored. A training ontology, the African Wildlife Ontology (AWO) was selected. It was extended using public lexical resources taken from the internet to include classes of common African plants and animals. This ontology was then visualised both as vectorspace embeddings and as a classical graph visualization. Participants were presented with one of four different knowledge graph visualizations: WebVOWL, OntoGraf, SquareVis and CircleVis and had to perform a specific information retrieval task. This task was to record as many African animals as they could find on the chart. The results are analyzed in terms of precision, recall, spam and average time. Although ultimately the results do not reject the null hypothesis, there is an opportunity for further research in the visualization of embeddings of knowledge graphs, especially for information retrieval.