Finished Theses
Bachelor's Theses
MetaStackVis: Visually-Assisted Performance Evaluation of Metamodels in Stacking Ensemble Learning
(thesis report)
Grading: Bachelor's Thesis
Area: Information Visualization / Visual Analytics / Machine Learning
Supervisor(s): Prof. Dr. Andreas Kerren, Angelos Chatzimparmpas
Active Student(s): Ilya Ploshchik
Content and Tasks: Stacking methods (or stacked generalizations) refer to a group of ensemble learning methods where several base models are trained and combined into a metamodel with improved predictive power. In particular, stacked generalization can reduce the bias and decrease the generalization error when compared to the use of single learning algorithms. To accomplish that, stacking enables the blending of different and heterogeneous algorithms and their instantiations with particular parameters, i.e., models. Other types of ensemble methods are bagging techniques, such as random forests, and boosting techniques, such as adaptive boosting or gradient boosting. A major difference between these ensemble methods is that stacking can use both bagging and boosting techniques in combination with simpler algorithms, stacked in different layers. It uses a meta-learner to aggregate the predictions of the last layer and obtain the best performance, which is absent in the other ensemble methods.We already have developed a visual analytics system, called StackGenVis, that assists users in dynamically adapting performance metrics, managing data instances, selecting the most important features for a given data set, choosing a set of top-performant and diverse algorithms, and measuring the predictive performance. In consequence, our proposed tool helps users to decide between distinct models and to reduce the complexity of the resulting stack by removing overpromising and underperforming models.
However, there is still remaining future work to be done. In detail, the student will investigate the impact of alternative metamodels on the predictive performance. Various data sets and machine learning models will be employed to address this challenge.
 |
Visualization of single-layer stacking ensembles with the use of a metamodel. Taken from here; there is also a video available. |
In this thesis, the student has to (1) familiarize themselves with the background information in ensemble learning visualization (e.g., by studying several research papers) and (2) develop a visualization tool in Python, using libraries such as Plotly. User studies will be performed to test the tool's usability.
Prerequisites:
- The motivation to learn and work hard investigating rewarding scientific problems.
- Programming skills in Python and JavaScript.
- Basic knowledge in D3.js, Plotly.js, and/or Scikit-learn libraries.
- Basic knowledge in information visualization, HCI, or machine learning is a plus.
- Basic knowledge in ensemble methods such as bagging, boosting, or stacking is a plus.
Using Leap Motion for the Interactive Analysis of Multivariate Networks
(thesis report)
Grading: Bachelor's Thesis
Area: Information Visualization / Visual Analytics
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Andreas Lif and Marcello Vendruscolo
Content and Tasks: The Leap Motion controller tracks hand movements and hand gestures. Typically it is used to interact with 3D virtual and augmented reality. Rarely it is used for navigation and interaction in 2D data visualizations, especially not for complex network visualizations.Aim of this work is to analyze the possibilities of Leap Motion to support the interaction with multivariate network visualizations (e.g., social or biological networks). The pros and cons of this technology should be analyzed and described. Finally, a software implementation based on JavaScript should be prepared that supports different navigation and interactions tasks, such as selecting and modifying network elements, etc.
 |
Image of a Leap Motion controller. |
In this thesis, the student has to analyze the possibilities of leap motion in context of interactive network visualization, incl. the definition of suitable gestures and their effects in the visualization. All these aspects shall also be implemented in JavaScript. In order to get some first validation results, a small and simple user study should compare the leap motion interaction with standard mouse interaction.
Prerequisites:
- Object-oriented programming in Java, JavaScript, eventually WebGL/OpenGL
- Knowledge in (2D/3D) computer graphics
- Basic knowledge in network visualization
Dynamic Co-authorship Network Analysis with Applications to Survey Metadata
(thesis report)
Grading: Bachelor's Thesis
Area: Data Mining / Social Network Analysis / Bibliometrics
Supervisor(s): Prof. Dr. Andreas Kerren, Dr. Kostiantyn Kucher
Active Student(s): Peter Johansson
Content and Tasks: Scientific co-authorship networks are based on the information about several co-authors publishing a paper together (for example, authors A and B publishing Paper 1 in 2019, and then authors B and C publishing Paper 2 in 2020). Analyses of such networks are important for the scientific community to identify core groups of researchers, find opportunities for future collaborations, etc. The analyses can even be generalized to other types of networks beyond scientific literature, for example, social networks.The co-authorship network data may include additional attributes attached to the nodes and edges, such as the number of papers published together by a given pair of researchers or the number of papers published in a specific year. The latter aspect opens up opportunities for modelling such networks as dynamic ones, which can be used to answer the following questions: (1) When did a collaboration between a given pair of researchers start or end, according to the data? (2) What are the temporal trends for a given connected component (a group of researchers) with regard to the number of members and publications? (3) What are the major events over time with regard to emergence, mergers, and splits of such connected components?
The specific application scenarios for such analyses include the metadata for several scientific literature surveys conducted by the ISOVIS group researchers. For example, our TextVis Browser data set includes publication metadata for more than 400 text visualization techniques described in scientific publications. The information about authors of these techniques could be extracted and analyzed—with the dynamic aspect (i.e., the publication year) in mind.
 |
Complete co-authorship network for TextVis Browser in May 2019 (taken from here). |
In this thesis, the student has to (1) review the existing work on the analysis of co-authorship networks and dynamic networks/graphs, (2) design and implement reusable and generalizable algorithms for dynamic co-authorship network analyses answering the questions above, and (3) apply the resulting algorithms to several survey data sets provided by the ISOVIS group. The validation of this work will include experimental results for the implementation (supplemented with further analyses in the existing network analysis tools such as Gephi), a discussion with the domain experts in visualization, and a critical discussion of the project outcomes and limitations.
Note: the student is not expected to design and implement any visualization approach in the scope of this thesis project, however, it would be eventually possible to continue this work in the scope of a follow-up Master's thesis project.
Prerequisites:
- Programming skills in Python, JavaScript, or Java (other programming languages are also an option)
- Knowledge in algorithms and data structures
- Basic knowledge in social network analysis, data mining, or information visualization is a plus
Comparison of Two Eye Trackers for the Visualization of Eye Tracking Data in Node-Link Diagrams
(thesis report)
Grading: Bachelor's Thesis
Area: Information Visualization / Visual Analytics
Supervisor(s): Prof. Dr. Andreas Kerren, Björn Zimmer
Active Student(s): Nazli Bilgic and Sofia Kiriaki Vulgari
Content and Tasks: The use of eye tracking devices is becoming more and more popular in the research field of information visualization. However, there are differences in the way how and why they are used. Mostly they are used for user studies in order to see what elements of a visualization are important for the user to solve specific tasks. Rarely they are used to influence the visualization itself, for example to change a visual representation (e.g., providing semantic zooming if a graphical element is in focus for a while).Aim of this work is to implement a prototype visualization that reacts on eye tracking data collected during a visualization session. A core problem to solve here is the identification of visualization elements when they are in the user focus. This should also safely work if the user interacts with the visualization (such as zooming or panning) or if he/she modifies the window parameters in which the visualization is displayed.
 |
Use of eye tracking data to evaluate visualizations. Taken from here. |
In this thesis, a software prototype for the dynamic adaption of visualizations based on eye tracking should be implemented that should be firstly applied within the network visualization system OnGraX. A small number of simple adaption possibilities should be developed too. In case the topic is chosen from two students, a comparison between two different eye tracking technologies in that context should be part of the final thesis report.
Prerequisites:
- Java, JavaScript (preferred), D3
- Knowledge in basic computer graphics
- Knowledge in basic information visualization
Visualization of Bags
(thesis report)
Grading: Master's or Bachelor's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Hui Wu
Content and Tasks: Multisets (also called bags) are used in many application fields, such as database systems or theoretical computer science. In contrast to a set, a multiset can contain the same element several times. This leads to a changed semantics regarding typical operations, such as union, intersection. etc. Whereas sets can be visualized with the help of Venn Diagrams, Euler Diagrams, or Edward's Venn Diagrams (cp. the figure below), the visualization of multisets is not straightforward.
 |
Edwards-Venn diagram for 6 sets. Taken from: Wikimedia Commons, 2008. |
In this thesis, an interactive visualization tool for multisets should be developed.
Prerequisites:
- Objekt-oriented programming, JavaScript, HTML 5
- Knowledge in (2D) computer graphics
Visual Analysis of Publication Networks
(thesis report)
Grading: Bachelor's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Jan Maushagen
Content and Tasks: We would like to visualize the co-authorship network of LNU with the help of Graph Drawing and InfoVis techniques and make the result available online.
 |
NodeTrix Representation. Taken from: NodeTrix: A Hybrid Visualization of Social Networks. N. Henry, J.-D. Fekete, and M.J. McGuffin. 2007. |
The aim of this thesis is the development of an interactive visualization tool that visualizes LNU's co-authorship relationships based on DIVA. A lot of visualization libraries are available and can be used to develop a web-based solution.
Prerequisites:
- Objekt-oriented programming
- Knowledge in computer graphics (2D), Graph Drawing, or Information Visualization
Visual Analysis of Swedish Research Council's Project Database
(thesis report)
Grading: Bachelor's Thesis
Area: Information Visualization and Visual Analytics
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Thorsten Haga
Content and Tasks: The Swedish Research Council offers access to an online database that provides information on all funded projects over the past years (currently 21,163 data items). Each funded projects has a set of attributes attached, for example, funding years, abstract, PI, institution, field, etc.
 |
An item from the project database of the Vetenskapsrådet (Swedish Research Council). Taken from: http://vrproj.vr.se/default.asp?funk=s. |
The aim of this thesis is the development of an interactive visualization tool that visualizes all data and provides a way to explore and analyze them. It should be possible to interactively filter out and to compare PIs, to focus on specific aspects, as well as to show abstracts if available. A lot of visualization libraries are available and can be used to develop a web-based solution.
Prerequisites:
- Objekt-oriented programming
- Database programming and web interfaces
- Good knowledge in computer graphics (2D) and Information Visualization
- Knowledge in Swedish would help
Creating Music Visualizations in a Mandelbrot Set Explorer
(thesis report)
Grading: Bachelor's Thesis
Area: Information Visualization
Supervisor(s): Björn Zimmer and Prof. Dr. Andreas Kerren
Active Student(s): Christian Knapp
Content and Tasks: Aim is to develop an application, which renders the Mandelbrot set and makes it possible to explore it. The application should provide different possibilities to navigate and to change the appearance of the set. Another goal is to develop an implementation of a music visualization, using the zoom into a Mandelbrot set. Zoom and colors should change accordingly to the music played. Therefore, it is necessary to analyze pitch and pace of the music.
 |
Initial image of a Mandelbrot set zoom sequence with a continuously coloured environment. Taken from: http://en.wikipedia.org/wiki/Mandelbrot_set. |
Prerequisites:
- Objekt-oriented programming in Java
- Knowledge in computer graphics (2D)
Visualization of Weather Data: Temperature Trend Visualization
(thesis report)
Grading: Bachelor's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren, Ilir Jusufi
Active Student(s): Jiayi Liu
Content and Tasks: There are a number of free on-line databases that offer access to different weather data and statistics accumulated over the years. These date can be used to analyze different patterns of weather conditions and could be helpful in understanding different meteorological phenomena. The aim of this thesis is to visualize such data. Users should be able to get an overview of the data and easily identify different patterns. This should be accomplished by developing a visualization tool.
 |
Avarage temperature in Vaxjo Daily temperatures since 1918 represented as matrix. Each row represents one year, and each pixel one day. Color is mapped to average daily temperature. |
- Objekt-oriented programming
- Knowledge in (2D) computer graphics, or Information Visualization
A Space-Filling Technique for the Visualization of Planar st-Graphs
(thesis report)
Grading: Bachelor's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren, Ilir Jusufi
Active Student(s): Yuanmao Wang
Content and Tasks: There are many cases in the real world where we need to model processes or data that have single source or/and single sink in their graph representation (st-graphs). One such example could be considered the PERT charts for project management. The task in this thesis is to develop a new space filing graph layout to overcome the problem of large graphs.
 |
An Example of st-graph representation. Taken from: Applications of Parameterized st≠Orientations in Graph Drawing Algorithms, Charalampos Papamanthou, Ioannis G. Tollis |
- Objekt-oriented programming
- Knowledge in (2D) computer graphics, Graph Drawing or Information Visualization
Visual Analysis of Author Impacts and Bibliometric Data
(thesis report)
Grading: Bachelor's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Björn Kostkevicius
Content and Tasks: Besides the visualization of co-authorship networks, we are also interested in the visual analysis of author impact information and so-called bibliometric data. An excellent example of a tool that provides such data is Harzing's Publish or Perish. This tool computes a set of different citation metrics, such as the well-known H-Index.
 |
Citation Map from ISI Web of Knowledge. Taken from: http://apps.isiknowledge.com. |
The aim of this thesis is the development of an interactive visualization tool that visualizes bibliometric data and provides a way to explore them for many authors. It should be possible to interactively filter out and to compare authors, to focus on specific aspects, as well as to show paper abstracts or articles if available. A lot of visualization libraries are available and can be used to develop a web-based solution. A later combination with the co-authorship visualization should be possible.
Prerequisites:
- Objekt-oriented programming
- Knowledge in computer graphics (2D) or Information Visualization
Visualization of Genealogical Data
(thesis report)
Grading: Bachelor's Thesis
Area: Information Visualization
Supervisor(s): Ilir Jusufi
Active Student(s): Jesús Miguel de la Fuente Bayo
Content and Tasks: Genealogical data can be presented in different forms, such as ancestor charts or family trees. Family trees usually present the oldest generations at the top and the newer generations at the bottom, opposite to ancestry charts which usually resemble more to a real life tree, being wider at the top (where newer generations are placed).With the expansion of different social networking websites such as Facebook, a lot of family members are getting "reconnected" on a virtual level. We could use the data provided by these websites and their users to build and visualize genealogical data. The task of this thesis is to develop an application (ex. Facebook Application) that will allow users to build and share visualization of genealogical data.
 |
Ahnenblatt Family Tree Taken from: Wikipedia, the free encyclopedia, 2009. |
- Objekt-oriented programming
- Knowledge in (2D) computer graphics, Graph Drawing or Information Visualization
Visualization of LNU's Publication Network
(thesis report)
Grading: Bachelor's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren, Ilir Jusufi
Active Student(s): Yu Chunmiao and Sun Wenyi
Content and Tasks: We would like to visualize the co-authorship network of LNU with the help of Graph Drawing and InfoVis techniques and make the result available online.
|
|
NodeTrix Representation. Taken from: NodeTrix: A Hybrid Visualization of Social Networks. N. Henry, J.-D. Fekete, and M.J. McGuffin. 2007. |
The aim of this thesis is the development of an interactive visualization tool that visualizes LNU's co-authorship relationships based on OPUS. A lot of visualization libraries are available and can be used to develop a web-based solution.
Prerequisites:
- Objekt-oriented programming
- Knowledge in computer graphics (2D), Graph Drawing, or Information Visualization
A Survey on Cloud Computing
and Prospects for Information Visualization
(thesis report)
Grading: Bachelor's Thesis
Area: Software/Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Hüseyin Öztürk
Content and Tasks: Asking Wikipedia, then "cloud computing is a style of computing in which dynamically scalable and often virtualized resources are provided as a service over the Internet. Users need not have knowledge of, expertise in, or control over the technology infrastructure in the 'cloud' that supports them." We would like to identify interesting research questions regarding visualization in this new upcoming field.
 |
Cloud Computing. Taken from: http://de.wikipedia.org/wiki/Cloud_Computing. |
The aim of this thesis is to give a comprehensive overview of cloud computing and to discuss possibilities where visualization could come into the play. In this way, this thesis is a more theoretical work, including a serious literature search and the capability to identify interesting research topics for the visualization community in the application field of cloud computing.
Prerequisites:
- Good fundamental knowledge in computer science
- Knowledge in computer graphics, graph drawing, or Information Visualization
Implementation of 3D-Kiviat Diagrams
(thesis report)
Grading: Bachelor's Thesis
Area: Information/Software Visualization
Supervisor(s): Andreas Kerren
Active Student(s): Guo Yuhua
Content and Tasks: Kiviat diagrams (or Star Glyphs, Star Plots) can be used to represent the attributes of a data element or the values of various data elements for a specific attribute. They are often used in statistics, but also in software visualization to visualize software metrics. Advantages lie in their simplicity. They can give a quick overview about the most important facts. However, they are often not suitable for visualizing the entire data set. A further drawback are overlapping polygonal lines, as the figure shows:
 |
A Kiviat diagram for visualizing software metrics. Taken from: Visualizing Multiple Evolution Metrics. Pinzger, Gall, Fischer, Lanza. SoftVis '05. |
In this thesis, disadvantages of existing visualizations of Kiviat diagrams shall be analyzed and then fixed through extension into 2.5D or 3D.
Prerequisites:
- Objekt-oriented programming
- Knowledge in computer graphics
Master's Theses
Task-Based Evaluation of Sentiment Visualization Techniques
(thesis report)
Grading: Master's Thesis
Area: Information Visualization Evaluation / Experimental Human-Computer Interaction
Supervisor(s): Dr. Kostiantyn Kucher, Prof. Dr. Andreas Kerren
External Co-Supervisor(s): apl. Prof. Dr. Achim Ebert (Technical University of Kaiserslautern, Germany)
Active Student(s): Samir Bouchama
Content and Tasks: Sentiment visualization is concerned with (interactive) visual representation of sentiments, emotions, opinions, and stances detected in textual data, for example, interactive charts or diagrams representing negative and positive opinions in online customer reviews or Twitter discussions. An online survey hosted by the ISOVIS group researchers at sentimentvis.lnu.se provides references to a wide range of sentiment visualization techniques categorized by the underlying data, designated technique tasks, and actual visual representation.While a variety of such techniques have been proposed over the years, the evidence about their effectiveness (i.e., whether they help the users to achieve their intended goals at all) and efficiency (i.e., whether they help to decrease the errors and required time) for specific tasks and problems is still lacking in many cases. Evaluation of information visualization and visual analytic approaches is an important research challenge, and further contributions are welcome, more specifically, within the area of sentiment visualization. The previous works describe (1) evaluation efforts for specific techniques and systems; (2) surveys and meta-analyses, such as the survey article published by the ISOVIS researchers in 2018; and (3) questionnaire results on the perceived usability of such techniques. The further steps involve a task-based evaluation implemented as a user study.
 |
Sentiment visualization survey statistics in April 2017 (taken from here). |
In this thesis, the students have to (1) familiarize themselves with the background information in sentiment visualization (e.g., by studying several survey articles), (2) review the existing work on task-based evaluation of information visualization approaches, and (3) design, conduct, and report the quantitative and qualitative results of a user study. The stages related to the experimental design and the discussion of the outcomes will be carried out with the support of domain experts in visualization.
Note: the students are not expected to design and implement any complex interactive visualizations in the scope of this thesis project—the main focus of the work will be on the evaluation of several basic visual encodings (i.e., several types of charts/diagrams) for sentiment visualization purposes. If a single student is carrying out the project, further support will be provided in this regard.
Prerequisites:
- Programming skills in Python, JavaScript, or Java (other programming languages are also an option)
- Basic knowledge in statistics
- Willingness to recruit and communicate with user study participants
- Basic knowledge in information visualization, HCI, or experimental design is a plus
- Basic knowledge in statistical data analysis with Python, R, Matlab/Octave/Scilab, etc. is a plus
Interactive Visual Analysis of Hypergraphs
(thesis report)
Grading: Master's Thesis
Area: Information Visualization / Visual Analytics
Supervisor(s): Dr. Kostiantyn Kucher, Prof. Dr. Andreas Kerren
Active Student(s): Ningrui Chen
Content and Tasks: A hypergraph is a generalization of a graph G = (V,E), where edges can connect any number of vertices, i.e., an edge e = (v1, ..., vk). Hypergraphs are difficult to draw, because they are more discussed using the terminology of set theory and not so much of graph theory. The image below shows a sample drawing of a small hypergraph by using the ONION tool developed in our group. Aim of this work is to re-implement the ONION tool in JavaScript, to fix mistakes and to extend the tool with additional interactive features. Those additions may include better edge bundling methods, node ordering to reduce the user's cognitive load, node/edge labeling, hypergraph comparisons, improved editing functionality, etc.
 |
Sample of a hypergraph drawn with the ONION tool |
In this thesis, a web-based interactive tool (JavaScript) for the visual analysis of hypergraphs should be developed according to the description given bove.
Prerequisites:
- JavaScript, Object-oriented programming in Java, eventually WebGL/OpenGL
- Knowledge in (2D/3D) computer graphics
- Knowledge in Information Visualization and Graph Drawing
Computational Analyses of Scientific Publications Using Raw and Manually Curated Data with Applications to Text Visualization
(thesis report)
Grading: Master's Thesis
Area: Data Mining / Natural Language Processing / Bibliometrics
Supervisor(s): Prof. Dr. Andreas Kerren, Kostiantyn Kucher
Active Student(s): Imran Shokat
Content and Tasks: Text visualization has become a prominent subfield of information visualization over the last decade. In order to provide an overview of the state of the art in text visualization to researchers and practitioners alike, we have developed a web-based visual survey called Text Visualization Browser which includes numerous manually curated entries corresponding to visualization techniques for raw textual data or text mining results. Each entry in this survey has been labelled by us according to a detailed hierarchical categorization consisting of 40 categories grouped into several higher-level categories, for instance, "Data" has nested subcategories "Source" and "Properties", and "Source" has low-level children categories "Document", "Corpora", and "Streams". Besides such categories, each entry contains several metadata fields such as title, publication year, and bibliographical reference.Several years ago, we conducted several analyses of the collected data for our survey, ranging from category distribution analysis to authorship statistics analysis and co-authorship network analyses, thus going into the direction of bibliometrics and scientometrics. Since then, the number of entries in our survey has grown from about 140 to about 400, which allows us to repeat our analyses and conduct other analyses relying on larger sample sizes. We are interested in the following questions: (1) What are the temporal trends related to individual categories in our survey data over time? (2) Is there any observable correlation between pairs of categories in our survey data? (3) Is there any correspondence between the category labels assigned to text visualization techniques manually and prominent concepts/topics described in the corresponding text visualization publications?
 |
Online survey browser (taken from http://textvis.lnu.se) and the underlying categorization of text visualization techniques (taken from here). |
In this thesis, the student has to conduct several types of statistical and text mining analyses using both labelled data (the manually curated survey data set) and unlabelled data (abstracts and full texts of the corresponding scientific publications). In order to answer questions (1) and (2), programming languages or environments such as Python, R, JavaScript, or Octave/Matlab should be used with the subsequent analysis of the results with spreadsheet software or visualization environments/libraries. In order to answer question (3), a software implementation should be created to retrieve and preprocess the text of scientific publications, conduct topic modelling, and compare the results to the labelled data set (Python or R are recommended for these tasks). The validation of this work will include experimental results for the topic modelling stage, a comparison to the existing meta-analyses of the text visualization subfield, and a discussion with experts in text visualization.
Prerequisites:
- Skills in Python, R, JavaScript, or Octave/Matlab
- Basic knowledge in natural language processing, especially topic modelling
- Basic knowledge in information visualization
Non-linear Dimensionality Reduction for the Visualization of Multivariate Networks
Grading: Master's Thesis
Area: Information Visualization / Visual Analytics
Supervisor(s): Prof. Dr. Andreas Kerren, Dr. Rafael Martins
Active Student(s): Hafed Mostafa Elgabroun
Content and Tasks: Our initial paper MVN-Reduce proposed a method for visualizing MVNs using DR by linearly combining the distances from attributes and connections. While the initial results were promising, this problem is far from solved. Most modern visualization-focused DR methods are non-linear, i.e., they adapt to different areas (clusters, neighborhoods, etc.) of the data set in different ways, as necessary. The main proposal for this thesis is:- To combine the distances from attributes and connections in a non-linear way: different points, neighborhoods or clusters may need different weights for this combination. How to obtain these non-linear weights will depend on the algorithm we use and the type of cost function we want to minimize.
- To allow users to interactively steer the underlying DR algorithm by exploring the map and choosing different parameter settings for different areas of the visualization.
- To propose and evaluate new visual quality measures for DR methods applied to MVNs, including a study on how these visual metrics can help guide the users' explorations.
- To systematically test different parameter settings, such as which distance measures should be used for both the attributes (euclidean, cosine-based, manhattan, etc.) and the connections (resistance, shortest path, random walk, etc.), and evaluate their impact on the visualization.
Prerequisites:
- The motivation to learn and work hard investigating rewarding scientific problems.
- JavaScript, WebGL, D3 and/or Processing.js
- Machine Learning (Dimensionality Reduction)
- Information Visualization especially with a focus on network visualization
Interactive Multiscale Visualization of Large, Multidimensional Datasets
(thesis report)
Grading: Master's Thesis
Area: Information Visualization / Visual Analytics
Supervisor(s): Prof. Dr. Andreas Kerren, Dr. Rafael Martins
Active Student(s): Kay Reemt Kühne
Content and Tasks: The use of interactive visualization techniques in Digital Humanities research can be a useful addition when traditional automated machine learning techniques face difficulties, as is often the case with the exploration of large volumes of dynamic -- and in many cases, noisy and conflicting -- textual data from social media. Recently, we developed StanceXplore, a new visualization for the interactive exploration of stance in social media. Our goal is to offer DH researchers the chance to explore stance-classified text corpora from different perspectives at the same time, using coordinated multiple views including user-defined topics, content similarity and dissimilarity, and geographical and temporal distribution. As a case study, we explore the activity of Twitter users in Sweden, analyzing their behavior in terms of topics discussed and the stances taken.Aim of this work is to significantly improve StanceXplore according to visualization & data scalability, temporal data analysis, consideration of real data streams and different levels of granularity.
 |
The StanceXplore tool for the visual exploration of stance in social media. Taken from here. |
In this thesis, the student will substantially extend the web-based visualization tool by using JavaScript and D3 or similar libraries. This extension will cover some technical aspects on the client and server side, but more importantly the inclusion of a visual display that is able to smoothly support granularity on multiple levels. In order to get some validation results, a small user study should prove the tool's usability.
Prerequisites:
- JavaScript, eventually WebGL/SVG/D3
- Knowledge in computer graphics
- Basic knowledge in information visualization
Updating the GaniFA Tool: GaniFA NG
(thesis report)
Grading: Master's Thesis
Area: Software Visualization
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Hüsamettin Yüksel
Content and Tasks: In context of the GANIMAL project, we designed an educational software system, called GANIFA. It is an electronic, HTML-based textbook on the theory of finite automata and their generation from regular expressions. GANIFA can be locally used as well as via the Internet. A short documentation of the applet parameter settings as well as features can be found here.
 |
One sample visualization step provided by the GaniFA tool |
In this thesis, the electronic textbook as well as the GaniFA applet should be updated and modernized. The applet should run with the newest JRE and the used graphical elements and layouts should get a more modern appealing. The old version of the HTML textbook shall also be revised and modernized.
Prerequisites:- Objekt-oriented programming (Java!)
- Knowledge in computer graphics
- Web Design (HTML, ...)
Evaluation of Network Comparison Approaches
(thesis report)
Grading: Masters's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren, Ilir Jusufi
Active Student(s): Hailelul Lakew Teshome
Content and Tasks: There are several different approaches for visual comparison of networks. Each approach may be intended for specific types of networks, or it can be domain specific.
 |
Network comparison using Juxtaposition + Superposition. Taken from: Visual Comparison for Information Visualization. M. Gleicher, D. Albers, R. Walker, I. Jusufi, C. D. Hansen, and J. C. Roberts. 2011. |
The aim of this thesis is to perform an user-centered evaluation of these approaches to find out the advantages and/or disadvantages of each approach.
Prerequisites:
- Knowledge in Graph Drawing, or Information Visualization
Visualization of Relationships in Clustered Text Data
(thesis report)
Grading: Masters's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren, Ilir Jusufi, Björn Zimmer
Active Student(s): Jiayi Liu
Content and Tasks: Visualization and analysis of large text collections plays an important role among different domains. Clustering a large dataset creates groups with similar attributes, but the user might want to find and explore hidden relationships between those groups. This thesis aims to provide an interactive tool for this task.
Prerequisites:
- Knowledge in Graph Drawing
- Knowledge in Information Visualization
ToDo Map
(thesis report)
Grading: Master's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren and Daniel Cernea
Active Student(s): Muddsair Sharif
Content and Tasks: Everyone has at a moment in time used a piece of paper to write down a ToDo list with the various tasks that one needed to complete the next day/s. Such ToDo lists are extremely useful in organizing and remembering tasks, as well as in time management. Also, powerful To Do list implementations are available for years now on different computing devices (e.g. PC, iPhone, tablets).But even if digitally supported, these lists are still mostly text-based, while much of the human sense-making and memory are visually oriented (shapes, colors, etc.). Furthermore, the main categorization of the tasks is based on time and date (calendar), type of activity, etc., while in general neglecting spatial positioning (e.g. a person that needs to remember to take a letter from his desk and bring it to the postal office the next day would generally imagine this entire process). To address this, we propose a new way of looking at digital ToDo lists that focuses on the location or locations where each task has to take place. We propose an implementation that marks places on maps and rooms models where various tasks have to be completed.
|
Conceptual designs of our idea. |
The student interested in this topic will have to develop an information visualization application in Adobe Flash (or Microsoft Silverlight), which should support the entire visualization and make it available on the web. The implementation would involve intuitive interaction and graph representations as presented in the images above.
Prerequisites:
- Objekt-oriented programming
- Adobe Flash (or Microsoft Silverlight)
Visualization of Gene Ontologies and Cluster Analysis Results
Grading: Master's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Vlad Aleksakhin
Content and Tasks: "The increasing complexity of genomic data presents several challenges for biologists. Limited computer monitor views of data complexity and the dynamic nature of data in the midst of discovery increase the challenge of integrating experimental results with information resources. The use of Gene Ontology enables researchers to summarize results of quantitative analyses in this framework, but the limitations of typical browser presentation restrict data access." [1].The general problem can be abstractly described. We have two different graphs: the gene ontology which is a Directed Acyclic Graph (DAG !) as well as a predefined clustering which is a tree (typically visualized as a dendrogram). The leaves of the dendrogram count 30.000 - 40.000 elements (genes). The aim is now to map the clustering to the gene ontology.
 |
An Example Visualization of a Gene Ontology. Taken from: [1] Visualization and analysis of microarray and gene ontology data with treemaps, Eric H Baehrecke, Niem Dang, Ketan Babaria, and Ben Shneiderman, 2004. |
The aim of this thesis is the development of a new visualization method that can visually map
the cluster tree onto the gene ontology DAG. The above mentioned example visualization with the help
of a treemap is not sufficient in some aspects.
Knowledge in biology is NOT needed for this work.
We also have a concrete idea how such a visualization can look like. Thus, the main task for
the student is the realization of that idea.
Prerequisites:
- Objekt-oriented programming
- Knowledge in computer graphics (2D/3D)
Visualization of Hypergraphs
Grading: Master's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Alev Koctas and Deniz Cakici
Content and Tasks: A hypergraph is a generalization of a graph G = (V,E), where edges can connect any number of vertices, i.e., an edge e = (v1, ..., vk). Hypergraphs are difficult to draw, because they are more discussed using the terminology of set theory and not so much of graph theory. The image below shows a sample drawing of a small hypergraph. Aim of this work is to develop a new kind of drawing (it can be a similar approach like the shown one) and to develop an interactive visualization tool that supports visual representation and intuitive interaction/navigation/exploration of large hypergraphs.
 |
Sample of hypergraph with E = {{v1,v2,v3}, {v2,v3}, {v3,v5,v6}, {v4}}. Taken from: Wikipedia, the free encyclopedia, 2008. |
In this thesis, an interactive visualization tool for hypergraphs should be developed.
Prerequisites:
- Objekt-oriented programming
- Knowledge in (2D/3D) computer graphics
Visualization of Network Centralities
(thesis report)
Grading: Master's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Harald Köstinger
Content and Tasks: Centrality analysis determines the importance of vertices in a network based on their connectivity within the network structure. It is a widely used technique to analyze network-structured data. A particularly important task is the comparison of different centrality measures within one network. In the life sciences centrality measures help scientists to understand the underlying biological processes and have been successfully applied to different biological networks. Central vertices in protein-protein interaction networks are often functionally important and the removal of such vertices is related to lethality. In metabolic networks metabolites with highest degree, i.e. with the highest number of neighbours, may belong to the oldest part of the metabolism and are main metabolites in well known pathways. Closeness centrality, a centrality that ranks vertices depending on the sum of their shortest paths to all other vertices is another method for identifying central metabolites in metabolic networks.
 |
An Example Visualization of centralities. Taken from: Visual Analysis of Network Centralities, Tim Dwyer, Seok-Hee Hong, Dirk Koschützki, Falk Schreiber and Kai Xu, 2006. |
The general problem is relatively simple: we have several instances of the same graph with ranked nodes. The ranking (i.e. the centrality) varies from graph to graph. Now, we would like to find a good visual representation for this problem. The aforementioned solution is not very good because of many overlaps etc. The aim of this thesis is the development of a new visualization that helps to analyze such network centralities.
Prerequisites:
- Objekt-oriented programming
- Knowledge in computer graphics (2D/3D)
Visualization of Sensory Perception Descriptions
(thesis report)
Grading: Master's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Mimi Prangova
Content and Tasks: Describing sensory perceptions, such as taste, is a challenging task. The visualization of these would help several groups of users to find out more. For a customer, the visualization of the taste of a specific wine or beer would help him/her to choose the right one (see here). On the other hand, for a language researcher, it could be interesting how humans talk about such sensory perceptions, what types of metaphorizations or lexical resources exist. Here, we are working together with Carita Paradis from the School of Humanities at VXU.
This thesis is focused on the language researcher side: here, we have access to a database of
sensory descriptions annotated with additional information. The aim of the thesis is the visualization
of possible correlations or patterns in the dataset to help linguists to better understand these
textual descriptions.
To reach that aim, an efficient visualization tool should be developed that supports an intuitive
detection of interesting structures as well as intuitive interaction possibilities.
Prerequisites:
- Objekt-oriented programming
- Basisc knowledge in computer graphics (2D)
The Network Lens
(thesis report)
Grading: Master's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Yang Dingjie
Content and Tasks: So-called Magic Lenses are one possibility to interact with complex visualizations, see the image below for example. A challenge in the visualization of complex and large networks is to integrate additional information into the drawing. One way to reach this aim, is to develop a magic lens that shows additional information if the user moves the lens over parts of the network. This can be done in several ways. The simple way is to show the information as separate visualization within the lens. Another possibility is to seamlessly intergrate it into the different network elements, for example, as visual node or edge attribute.
 |
A magic lens! This lens (mouse-controlled rectangle) allows a better selection of the single parts of the drawing. |
The aim of this theses is the development of a tool that supports to embed additional information into network drawings with the help of the magic lens approach.
Prerequisites:
- Objekt-oriented programming
- Base knowledge in computer graphics (2D)
Visualization of Wine Attributes
(thesis report)
Grading: Master's Thesis
Area: Information Visualization
Supervisor(s): Prof. Dr. Andreas Kerren
Active Student(s): Yuanxun Mei
Content and Tasks: Describing sensory perceptions, such as taste, is a challenging task. The visualization of these would help several groups of users to find out more. For a customer, for example, the visualization of the taste of a specific wine or beer would help him/her to choose the right one. For a language researcher, it could be interesting how humans talk about such sensory perceptions, what types of metaphorizations or lexical resources exist, etc. Here, we are working together with Carita Paradis from the School of Humanities at VXU.The example visualization below shows a visualization of a book, i.e., literature. This idea of developing a fingerprint could be an excellent starting point for further investigations into this very interesting topic from a customer point of view.
 |
An example of the so-called literature fingerprinting. Taken from: D. Keim, D. Oelke: Literature Fingerprinting: A New Method for Visual Literary Analysis, IEEE Symposium on Visual Analytics and Technology (VAST 2007), 2007. |
The offered thesis is focused on the customer side, i.e., the general idea should be clear from our daily life. The aim of this thesis is the development of a tool for wine fingerprinting. It should be possible for a wine seller to generate a fingerprint for a specific wine with the help of this tool. On the other hand, the customer should be able to use this tool, for example, via a web interface, to produce wine fingerprints for his/her own tastes. Then it is possible to generate individual fingerprints for specific kinds of wines or opportunities, such as a preferred wine for diner or as a companion for a cigar. Finally, the customer could compare his/her own fingerprints with those given for each wine in a wine store to find a suitable one.
Prerequisites:
- Objekt-oriented programming
- Basisc knowledge in computer graphics (2D)
Visualization of Text Duplicates in Documents
(thesis report)
Grading: Master's Thesis
Area: Information/Software Visualization, Visual Analytics
Supervisor(s): Andreas Kerren
Active Student(s): Chao Wang and Han Pan
Content and Tasks: The visual analysis of text duplicates is important for a lot of different application scenarios. One example is the discovery of software clones: finding and analyzing of those lead to a better software quality and avoid bugs. Another example is the discovery of plagiarism in different documents: a visualization tool that allows us to find similar text parts in different documents could help us in this matter.Often, documents are hierarchical organized, such as Java code or XML documents. The intended visualization tool should show such hierarchies too. The screenshot example below shows the old SeeSoft visualization tool displaying differences between two software systems. Here, the package structure is clearly to see.
 |
An example visualization (SeeSoft) for showing differences between two systems. Taken from: S.C. Eick, J.L. Steffen, E.E. Sumner Jr. Seesoft - A Tool for Visualizing Line Oriented Software Statistics. IEEE Transactions on Software Engineering, vol. 18, no. 11, pp. 957-968, Nov. 1992. |
At first, we will focus on the analysis of software clones because we have access to a good data set (including static analysis results) from the software engineering group at VXU. The idea is to show a rough overview visualization at first in order to discover where duplicates occur (like a fingerprint). Then, the visualization should become more finegrained step-by-step (so-called semantic zoomimg) depending of the user interaction and the underlying data set. At the lowest level, the tool should show equal/similar text parts directly.
Prerequisites:
- Objekt-oriented programming
- Basisc knowledge in computer graphics (2D)
Evaluation of a Hypermedial Learning System for Non-linear Explanation of Algorithms (in German)
Grading: Master's Thesis
Area: Software Visualization
Supervisor(s): Prof. Dr. Andreas Kerren, Prof. Dr. Stephan Diehl
Active Student(s): Rebecca Wessel
Content and Tasks: Most systems designed to teach algorithms using visualization and animation techniques have not proved to be educationally effective. To satisfy this aim, some recently built systems use a hypermedia environment to provide knowledge and context to explain algorithms. Currently, we develop a system called Structured Hypermedia Algorithm Explanation (SHALEX), which provides several novel and important features. In particular, our hypermedia environment can reflect the structure of an algorithm.
|
Abstract description of an algorithm within the SHALEX system. |
In this thesis, the graphical interface on the learner level should be improved. Currently, we only have a simple dialog box as GUI. After testing the entire tool, an evaluation with students together with performance tests will be performed in order to prove the usability and efficiency of our approach.
Prerequisites:
- Objekt-oriented programming (Java)
- Base knowledge in statistics (expectation, random variables, etc.) for the evaluation
- Interest in design and human-computer interaction
Dynamic and Static Approaches for Glyph-based Visualization of Software Metrics
(thesis report)
Grading: Master's Thesis
Area: Software Visualization
Supervisor(s): Andreas Kerren
Active Student(s): Raja Majid Mehmood
Content and Tasks: The visualization of so-called software metrics (e.g., lines of code, code coverage, ...) is very important for the better understanding of software and for increasing software quality. There are a lot of examples of visual representations, such as treemaps or Kiviat diagrams (star plots). Our aim is to develop a new kind of visual representation that can be embedded in larger visual structures like a graph. Thus, it should be space-filling, i.e. it should only need a specific amount of space in the 2D plane.The extension to dynamic visualizations (animations) offers new possibilities to discover patterns and correlations, not only by using a further dimension (time).
The aim of this theses is the development of a new static and dynamic visualization for software metrics.
Prerequisites:
- Objekt-oriented programming
- Base knowledge in computer graphics (2D)
- Basics in Software Engineering
Content Visualization of GeoAudio Notes
(thesis report)
Grading: Master's Thesis
Area: Information Visualization
Supervisor(s): Andreas Kerren, Marcelo Milrad
Active Students: Ilir Jusufi and Lulzim Junuzi
Content and Tasks: The total population of GPS-enabled location-based services (LBS) subscribers is constantly increasing. This implies the research of the new possibilities of visualizing geospatial data produced by these mobile devices.The aim of this thesis is the development of a tool that visualizes the content of voice notes that will be placed in maps using GPS coordinates, and visualizing the semantical, temporal, and spatial relations between the notes.
Prerequisites:
- Objekt-oriented programming
- Base knowledge in computer graphics (2D)