Konversatorium on Friday, August 7, 2020 - 10:30

Concept maps are a method for the visualization of knowledge and an established tool ineducation, knowledge organization and a variety of other fields. They are composed of concepts and interlinked relations between them and are displayed as a node-link diagram. Concept map mining is the process of extracting concept maps from unstructured text. The three approaches to mine concept maps are: manually, semi-automatic or fully automatic. A fully automatic approach cannot mirror the mental knowledge model, which a user would transfer to a manually created concept map. The manual process is often perceived as tedious and inefficient, limiting a wide-range application of concept maps.

This thesis presents a semi-automatic concept map mining approach that tries to bridge the gap between all manual construction and fully automatic approaches. The advantage of this approach is that the users still have control over how their concept map is constructed, but are not impeded by manual tasks that are often repetitive and inefficient. The presented approach is composed of an automatic text processing part, which extracts concepts and relations out of an unstructured text document and is powered by state-of-the-art natural language processing and neural coreference resolution. The second manual concept map creation part allows the creation of concept maps in a user interface and presents the extracted concepts and relations as suggestions to the user.

In a user study, an implemented prototype of the proposed semi-automatic concept map mining approach was evaluated. Manual gold standard concept maps that were created by the users and concept maps created by a fully automatic tool were compared to concept maps that were created with the prototype, proving the usefulness of the process. Results show that created concept maps with the semi-automatic prototype are significantly more similar to the gold standard than the ones created by the fully automatic tool. Additionally, considerably improved efficiency in creation duration and user satisfaction could be observed in comparison to the manual creation of the gold standard maps.