Springe direkt zu Inhalt

Bachelor Thesis Defence: "Evaluating Information Landscapes for Interpretation of Research Topic Embedding by Non-technical Stakeholders" by Tal Levy (Dec 17, 2020)

News from Dec 08, 2020

December 17th, 2 pm, the Thesis Defense: "Evaluating Information Landscapes for Interpretation of Research Topic Embedding by Non-technical Stakeholders" by Tal Levy will take place at HCC Lab.

Abstract

The exponential rise in capability and popularity of machine learning algorithms has brought about advancements in many areas of life. From recommendation engines to weather prediction models, the ability to find significant patterns in large databases and produce data-driven decisions has changed the way many non-technical professionals approach solving real world problems. An often cited barrier to wider acceptance of machine learning algorithms is the lack of accessibility for non-technical stakeholders. As a way to increase the interpretability by non-technical experts, known webs of significance can be used to convey information that otherwise might not be obvious. One such common knowledge metaphor is the landscape metaphor. In this thesis we evaluate the possible benefits of spatialization and application of the landscape metaphor on a non-geographic data set. Using the GEPRIS database, a collection of projects funded by the DFG (Deutsche Forschungsgemeinschaft), we implement a pipeline which converts the body of research into a 2D geographic map using labels, color palettes and shading in order to conjure the landscape metaphor in the eyes of observers. This representation is then evaluated using a think aloud pilot study conducted on non-technical stakeholders in different academic fields. During the evaluation process our participants were tasked with clustering these maps according to academic context. Beyond clustering the landscape representation, participants were asked to cluster maps comprising of points, labels and label weights alone. Using the results of these tasks and the process of contrasting and comparing the clustering process of these different representations, we will attempt to evaluate the advantages of said metaphor in helping with the interpretability of 2D representations of large data sets.

First assessor: Prof. Dr. Claudia Müller-Birn
Second assessor: Prof. Dr. Lutz Prechelt

Location: WebEx (If you want to join the thesis defence, please contact Phil Wernberger.)

3 / 100