Bachelor- und Masterarbeiten Sommersemester 2023

Temporal Ordinal Motifs in Topic Models

Topic models are, often, dimension reduction techniques for large corpora of textual documents. A central aspect to these models is that they allow for text based explanations of the dimensions in the reduced space. A novel technique, called ordinal motifs, interpret and visualize these dimension hierarchically with respect to (ordinal) substructures of standard shape. With your work, you extent this technique towards ordinal motifs over time, develop visualization techniques, and show their applicability in a practical setting.

Informationen: Johannes Hirth

Ordinal Motifs in Hierarchical Topic Models

Topic models are, often, dimension reduction techniques for large corpora of textual documents. A central aspect to these models is that they allow for text based explanations of the dimensions in the reduced space. A novel technique, called ordinal motifs, interpret and visualize these dimension hierarchically with respect to (ordinal) substructures of standard shape. With your work, you extent this technique towards hierarchical topic models, define hierarchical motif structures, develop visualization techniques, and show their applicability in a practical setting.

Informationen: Johannes Hirth

Network Motifs in Topic Flow Networks

In scientometrics, scientific collaboration is often analyzed by means of co-authorships. An aspect which is often overlooked and more difficult to quantify is the flow of expertise between authors from different research topics, which is an important part of scientific progress. With the Topic Flow Network (TFN) a graph structure for the analysis of research topic flows between scientific authors and their respective research fields was proposed. With your work, you identify and interpret substructures that are integral to this network.

Informationen: Johannes Hirth

Formal Concept Analysis mit Attribut und Objektordnungen

In dieser Arbeit untersuchen Sie, inwiefern sich die Theorie der formalen Begriffsanalyse auf den Fall übertragen lässt, dass wir eine lineare Ordnung auf den Attributen und den Objekten vorliegen haben.

Das Ziel ist es, die in der FCA üblichen Ideen (Begriffe, Implikationen etc.) auf solche Datensätze zu übertragen und die Theorie mit Echtwelt-Datensätzen zu evaluieren.

Informationen: Dominik Dürrschnabel

Implications in Conceptual Scaling

One way of computing dependencies in data set are implications. To extract implications from data sets, we first have to interpret the data on the ordinal level via a method called conceptual scaling. The implication that we find in the scaled data set can have two origins. The first are dependencies in the many-valued data set and the second are artifacts from the scaling process. With your work you develop a method to analyze these sets of implications separately.

Informationen: Johannes Hirth

Decomposition of Concept Lattices

Conceptual structures are great hierarchical tools to analyze complex relations between data point. Recent approaches focus on identifying ordinal sub-structures of concept lattices that have specific shape, e.g., chains, cubes, cycles etc. The sub-structures are then used to derive highler level relations between data point or to explain the hierarchical structure. With your work, you study how this approach can be used to decompose concept lattices into sub-structures.

Informationen: Johannes Hirth

Stabilität von Formalen Kontexten

Wir nennen einen formalen Kontext stabil, wenn sich beim Setzen oder Entfernen jedes Kreuzes die Größe des Begriffsverbandes nicht verkleinert. Untersuchen Sie Echtweltdaten auf Stabilität und untersuchen Sie, inwiefern sich die Stabilität als Bewertungsmaß für intrinsisch sinnvolle Daten eignet.

Informationen: Dominik Dürrschnabel

Author Identification based on Paper Citations

Several venues use the double-blind review process to evaluate submitted research articles. We would like to understand how the citations used in the new paper give access to the identity of the anonymous author(s). An important part of your work would be the evaluation of previously proposed techniques [1] on new raw [2] or processed [3] datasets. Of course, there is also the possibility to apply newer approaches and your own ideas.
Most (if not all) of the programming will be done in Python.

Inquiries: Tobias Hille

Sanity Checks for Conceptual Views on Neural Networks

Conceptual views provide a new method to interprete the latent representations of a neural network. In this work, you compare the sensitivity of conceptual views to weight randomizations for different architectures.

Informationen: Johannes Hirth

Investigate Formal Conceptual Views in Neural Networks

Mit Formal Conceptual Views ist ein neues Modell zur Erklärung von Neuralen Netzwerken vorgestellt worden. Ziel dieser Arbeit ist es die Methode auf weitere Modelle und Daten Domänen anzuwenden und die Formal Conceptual Views mit weiteren Methoden der symbolischen Wissensverarbeitung zu analysieren.

Informationen: Johannes Hirth

Intrinsic Triangulation of Loss Landscapes of Neural Networks

Recent work investigating geometry and topology of loss landscapes of neural networks revealed interesting properties regarding connectivity of clusters of local minima [1], [2]. In this work you will try to apply advances in computation intrinsic triangulation for 3d rendering [3] towards those surfaces (or reasonable hyperplane projections). One possible motivation for this is the change for an ability to construct geodesic paths on the constructed approximation.
Most (if not all) of the programming will be done in Python.

Inquiries: Tobias Hille

Extracting Hypergraphs from Traffic Networks

Hypergraph datasets are well known in the context of author networks and knowledge graphs. However, hypergraphs can also be found in other domains, such as traffic networks. In this work, you should search for existing datasets from this domain and provide tooling to collect data from public web endpoints. A second part will deal with characterising different ways of interpreting and formatting the data as hypergraphs.
Most (if not all) of the programming will be done in Python.

Inquiries: Tobias Hille

Core Numbers in Bipartite Networks

Core numbers are efficient valuations for nodes in networks. They are a measure to describe the structural integration of nodes. In this work, you should characterize and study core numbers for bipartite graphs.

Informationen: Johannes Hirth

Community Detection in WikiData zur Datensatzgenerierung

Knowlegde Graphs wie WikiData enthalten sehr viel Wissen, das im Bereich der künstlichen Intelligenz in vielen Anwendungen eingesetzt werden kann. Der Umfang an Informationen ist aber auch ein Problem für viele Anwendungen. Ihre Aufgabe besteht darin, Methoden der Community Detection in sozialen Netzwerken auf die Struktur von WikiData zu übertragen. Des Weiteren sollen Sie untersuchen, wie diese Methoden genutzt werden können, um kleinere Teil-Datensätze aus WikiData zu extrahieren.

Informationen: Johannes Hirth