Gavagai presents a poster on Semantic Topology at this week’s ACM sponsored CIKM conference on information and knowledge management, in Shanghai. The poster is based on recent bachelor’s theses by KTH students Martin Bohman, Ariel Ekgren, Gabriel Isheden, Emelie Kullmann, and David Nilsson, supervised by Jussi Karlgren. The experiments show the usefulness of using computational topology as a tool for working with semantic spaces.
This publication marks the first public steps of moving from a black-box view of learning language models to a more explicit knowledge representation. The aim is to continue the development of semantic space models towards models that not only learn incrementally and handily but that also are aware of if they know stuff or not, and what of what they know is important or less important.