Master

Titel: Predicting the Behavior of Drosophila Melanogaster

{web_name}e Masterarbeit wird in Kooperation mit der Neurogenetik an der UR vergeben:

Behavioral variability explains why under similar circumstances, individuals will generate different actions. The fruit fly Drosophila can show random-like behavior. The goal is to identify and characterize the neural generators of this random-like behavior in the fly and its consequences. For this purpose, we analyze the behavior of Drosophila in flight and walking and try to predict and characterize it with machine learning algorithms.

Weitere Informationen zum Projekt unter: http://lab.brembs.net/biofoo_project/hide-if-you-cant-fly-phototaxis-change-after-wing-clipping-in-drosophila/

Kontaktperson: Prof. Dr. Bernd Ludwig

Titel: Miss-information amongst refugees

Miss-information, half-truths and fake news have received considerable attention in the media in recent times. Most of the research attention in this space has focused on social-media and how rumors and untruths are spread. However, false information can be spread in many ways.

Taking the refugee population as a case study, the aim of this project is to determine whether there are commonly held beliefs that are formed within refugee communities in Germany that are not true and understand how are these formed.

Kontaktperson: PD Dr. David Elsweiler

Titel: Information sharing / withholding amongst refugees - a study in the mould of Elfrede Chatman

Elfrede Chatman can be considered as one of the most influential information scientists. Her work studied information poverty and in particular she used ethnographical methods to understand the information landscape and behaviour of particular social groups, including university janitors, elderly women in care homes and the homeless. This project would continue her work to understand another underprivileged demographic - political refugees in Germany.

Skills and interests: qualitative research

Kontaktperson: PD Dr. David Elsweiler

Titel: Information seeking strategies of refugees

The refugee crisis has brought increasing numbers of refugees to Germany. When they arrive here they are encountered with many hurdles. How do things work? Where I can I stay? How long can I stay? Where can I get support? Cultural differences mean that the answers are often puzzling and all of this happens in a foreign language. This project will attempt to discover how refugees attempt to find answers to these kinds of questions. More specifically it will try to answer What kinds of information needs refugees have? What information resources do they have, which ones do they use hand how? What problems do they encounter?

Kontaktperson: PD Dr. David Elsweiler

Titel: Analysing Trends in Online Supermarket Data

Government health initiatives often try to encourage home cooking as a means to promote healthier diets [1,3]. Nevertheless past studies have shown that common sources of cooking inspiration, such as recipes published by celebrity chefs tend to be unhealthier in many ways [2]. These kinds of studies are typically based on very small hand picked samples. In Howard's work a sample of recipes from celebrity cookbooks with a small sample of ready meals from UK supermarkets. While such studies are insightful, they do not show the full picture.

This project would involve crawling data from the website http://www.mysupermarket.co.uk/ to determine a more accurate and complete dataset from which to investigate the healthiness of ready meals. When the dataset has been created data mining approaches can be employed to answer a number of questions, which previous studies have not yet been able to answer. These include:

• How healthy are ready meals?
• Does this vary across stores?
• Is there a relationship between price and healthiness?
• Is there a relationship between popularity and healthiness?

References

Department of health. change4life marketing strategy. 2009. available athttp://www.nhs.uk/Change4Life/supporter-resources/downloads/Change4Life_Marketing%20Strategy_April09.pdf. last accessed on20.6.2016.

Howard, Simon, Jean Adams, and Martin White. "Nutritional content of supermarket ready meals and recipes by television chefs in the United Kingdom: cross sectional study." BMJ 345 (2012).

Usda.cookmoreoftenathome.2011.availableathttp://www.choosemyplate.gov/weight-management-calories/weight-management/better-choices/cook-home.html. last accessed on20.6.2016

Kontaktperson: PD Dr. David Elsweiler

Titel: Studying how online recipes are used in practice

Our recent work has demonstrated that users of online food portals generally prefer less healthy meals and their choices can be influenced by several factors including presentation, algorithms and personal preferences. That being said, just because online recipes are interacted with online does not mean that they are actually prepared and consumed by users. This project will investigate how online recipes are actually used in practice and more specifically, attempt to answer the following research questions:

• When and why do people use online recipes? (for special occasions? to try something new? because they cannot cook without them?)
• How are they used? Do people tend to follow the ingredients and instructions exactly or are they just used as inspiration?
• When do users deviate and why?
• In which ways and to what extent does the finished product vary from what is described in the recipe?

The results are important not only because they help us understand people’s behaviour, but also to help us estimate to what extent interactions with online recipes can be seen as a proxy for eating a meal with particular nutritional content.

Potential approaches: diary study, interview, observation.

Kontaktperson: PD Dr. David Elsweiler

Food ordering system for a sports club

Although participating in sport is typically associated with a healthy lifestyle, many of the community sports clubs in Germany provide particularly unhealthy food to the members. This project would involve developing an online food ordering system allowing members to order their food in advance. Not only would this make life easier for both the members and the restaurant, it would also provide a platform to understand food recommendations and potential behavioural change.

Kontaktperson: PD Dr. David Elsweiler

Website / Search engine to communicate and explore the geographical trends in health statistics and how this relates to the online food recipes with which people in those regions interact.

This project will involve indexing online recipes to make them searchable and at the same time visualising the trends found in recent publication.
Here is an example of such a site as means of inspiration.

Kontaktperson: PD Dr. David Elsweiler

Titel: Detecting social media posts on controversial topics

The content posted on social-media is extremely diverse, ranging from posts documenting personal activities to opinions and statements of facts. Some of these facts are mundane and undisputed whereas others are extremely controversial. The aim of this project is to identify controversial tweets posted by German politicians. The student will develop and test automatic methods of detection using a large collection of tweets collected by elected politicians in Germany, which is being collected as part of an i:imsk collaboration.

This will involve some technical and programming skills as well as knowledge of statistics.

Kontaktperson: PD Dr. David Elsweiler

Titel: Studying images posted by politicians on Twitter

There are increasing numbers of studies investigating the social-media behaviour of politicians. These have applied both qualitative and quantitative approaches, demonstrating that politicians have diverse aims and behaviours, but in doing so exhibit many key cultural phenomena known from cultural studies. Many of the Tweets politicians post contain images. The aim of this project is to understand what kinds of images are posted, in which situations and why they might want to do this. The student will be provided with access to a large collection of tweets collected by elected politicians in Germany, which is being collected as part of an i:imsk collaboration.

Kontaktperson: PD Dr. David Elsweiler

Titel: Games with sensors to encourage children in their movement and increase coordination skills.

According to the WHO, 41 million children under the age of 5 were overweight or obese in 2014 (WHO2014). However, obesity is preventable with a combination of healthy diet and adequate physical exercise. The aim of this project is to devise and develop interactive games for children to play, which will encourage them to be more active.

This project will require and interest in / and knowledge of programming and embedded systems (via raspberry pi).

Kontaktperson: PD Dr. David Elsweiler

Titel: Colour-based Food recommender

The colourfulness of food plays a key role in food choice by influencing taste thresholds, sweetness perception, food preference, pleasantness, and acceptability [1]. A large body of research has shown that changing the hue or intensity of the colour of food items can exert a sometimes dramatic impact on the expectations and subsequent experiences of consumers [2]. There is also evidence suggesting meals with broad range of colours tend to more healthy.

While it makes sense that the image associated with a recipe will be important in determining how it is perceived and rated, in the recommender systems community no research to date has investigated this in detail. Using large datasets collected via popular Internet food portals (allrecipes.com and kochbar.de), this project will study the relationship between the colourfulness of food images and perception of the food and will investigate the feasibility of using colour information to provide recommendations. Potential research questions include:

Does the colour profile of a recipe correlate with the healthiness of the recipe? Does the colour profile of a recipe correlate with the popularity of the recipe? Is it possible to predict user recipe ratings based on the colour profiles of previously liked recipes?

[1] Clydesdale, Fergus M. "Color as a factor in food choice." Critical Reviews in Food Science & Nutrition 33.1 (1993): 83-101. http://www.ncbi.nlm.nih.gov/pubmed/8424857

[2] Spence, Charles. "On the psychological impact of food colour." Flavour 4.1 (2015): https://flavourjournal.biomedcentral.com/articles/10.1186/s13411-015-0031-3

[3] http://blog.revolutionanalytics.com/2015/03/color-extraction-with-r.html

Kontaktperson: PD Dr. David Elsweiler

Titel: "Genomdaten online"

(Kooperation mit dem Lehrstuhl für Genetische Epidemiologie)

Google und andere gro?e Konzerne investieren hohe Betr?ge in die Genetik. Ein aktueller Trend ist, dass sich Personen ihr eigenes Genprofil auf einem speziellen Chip "speichern" lassen k?nnen. Dafür gibt es immer mehr Internet-Plattformen. Die gesellschaftlichen Konsequenzen dieser Entwicklung sind unabsehbar.

Ziel der Arbeit ist es, Online-Portale im Internet ausfindig zu machen, die anbieten, die DNA eines Nutzers zu analysieren und/oder Genomdaten von Personen online verfügbar zu machen.

Dabei sind folgende Leitfragen relevant:

• Wer sind die Anbieter dieser Portale?
• Welche Absichten verfolgen sie?
• Sind die genomweiten Daten mit Namen oder anderen identifizierenden Merkmalen versehen?
• K?nnen die genomweiten Daten von Dritten gespeichert werden?

Zur Bearbeitung sind (automatisch) geeignete Suchanfragen zu generieren, durchzuführen und auszuwerten. Identifizierte Seiten k?nnen mit fingierten Anfragen getestet werden. Alle Anfragen sind für die sp?tere Auswertung zu protokollieren.

Abh?ngig davon, welche und wie viele Informationen gesammelt werden k?nnen, besteht die Evaluation des Vorgehens in einer qualitativen oder quantitativen Analyse der Portale und der zur Verfügung gestellten Informationen.

Kontaktperson: Prof. Dr. Bernd Ludwig

K?nnen Low-Level Image Features Indoor-Salience vorhersagen?

Im Rahmen eines Seminars wurde ein Datensatz mit insgesamt ca. 7000 Bewertungen von Indoor-Objekten erhoben. Jede Bewertung umfasst drei Fragen zur Messung der Gesamtsalienz, die Kattenbeck (2016) entnommen wurden.

Die Masterarbeit soll der Frage nachgehen, ob die Fragebogen-basierten Salienzwerte mit low-level Image Features (z.B. lokale Farb?nderungen etc.) hinreichend gut vorhergesagt werden k?nnen.

Zur Beantwortung dieser Frage sollen verschiedene, state-of-the-art Machine Learning Verfahren herangezogen werden.

Literatur:

• Kattenbeck, Markus (2016): Empirically Measuring Salience of Objects for Use in Pedestrian Navigation.
• Dominguez et al. (2017):?Comparing Neural and Attractiveness-based Visual Features for Artwork Recommendation, verfügbar unter: arXiv.org/pdf/1706.07515.pdf

• Itt &?Koch (2001): Computational modelling of visual attention, verfügbar unter www.nature.com/nrn/journal/v2/n3/full/nrn0301_194a.html

Kontaktperson: Prof. Dr. Bernd Ludwig

Identifikation globaler Landmarken am Campus der Universit?t

Globale Landmarken, also solche die abseits der Route liegen, unterstützen die Orientierung w?hrend der Navigation und f?rdern die Entwicklung einer kognitiven Karte. So kann man beispielsweise auf die ?Uni-Kugel“ in viele Situationen gut versweisen, da sie ein bekannter Treffpunkt ist. Die Integration globaler Landmarken in Routenanweisungen insbesondere für den Innenbereich ist das übergeordnete Ziel der Arbeit.

Im Rahmen des Seminars URWalking wurden bereits mehrere Landmarken am Campus erhoben. Die Arbeit soll sich mit der Analyse dieser Objekte hinsichtlich

• Art (fl?chig, lokal, etc)
• Sichtbarkeit vom Standpunkt der Anweisung
• Position innerhalb/au?erhalb eines Geb?udes
• Abstand des Objekts zum Standpunkt besch?ftigen.

Hierzu müssen diese Daten zun?chst annotiert werden. Daraufhin sollen unterschiedliche Eigenschaften globaler Landmarken identifiziert werden, die über die Position relativ zur Route hinausgehen.

Literatur:

Anacta, V. J. A., Wang, J., & Schwering, A. (2014). Routes to Remember: Comparing Verbal Instructions and Sketch Maps. In Connecting a Digital Europe Through Location and Place (pp. 311-322). Springer International Publishing.

Lovelace, K. L., Hegarty, M., & Montello, D. R. (1999, August). Elements of good route directions in familiar and unfamiliar environments. In International Conference on Spatial Information Theory (pp. 65-82). Springer Berlin Heidelberg.

Schwering, A., Li, R., & Anacta, V. J. A. (2013). Orientation information in different forms of route instructions. In Short Paper Proceedings of the 16th AGILE Conference on Geographic Information Science, Leuven, Belgium.

Kontaktperson: Prof. Dr. Bernd Ludwig

Titel: Auswertung der URWalking-Logdaten

Unsere URWalking-Webapp speichert seit einigen Monaten alle Anfragen samt Geokoordinaten etc. in einer Datenbanktabelle. Die Tabelle ist ziemlich gewachsen und dementsprechend unübersichtlich.

Um unsere Navigation besser an die Bedürfnisse der Nutzer anpassen zu k?nnen, w?re nun beispielsweise interessant:

• Welche Nutzer fragen Routen ab (sind das nur Lehrstuhlmitarbeiter oder auch Studenten oder andere Mitarbeiter)?
• Welche Ziele werden abgefragt und wie k?nnen wir diese Daten nutzen, um unser System sinnvoll zu erweitern?
• Welche Probleme treten bei der Start-/Zieleingabe auf?
• Wie interagieren Nutzer mit dem System?

N?tig w?ren Grundkenntnisse in Datenbanken (PostgreSQL) und Erfahrung mit einer beliebigen Programmiersprache.

Kontaktpersonen: PD Dr. David Elsweiler

WLAN Heatmaps (in Kooperation mit dem Rechenzentrum)

Für die URWalking-Android-App soll eine Funktion entwickelt werden, die Nutzern anzeigt, wo sie die beste WLAN-Qualit?t haben.

Kontaktperson: Prof. Dr. Bernd Ludwig

Bewegungslogging über WLAN

Für die URWalking-Android-App soll eine Funktion entwickelt werden, die über das Speichern von empfangenen WLAN-Signalen aufzeichnet, welche Hotspots Nutzer passieren. Daraus sollen pr?ferierte Routen abgeleitet werden.

Kontaktperson: Prof. Dr. Bernd Ludwig

Weitere Themen:

• Vorhersage von Landmarken
• Generierung kontextabh?ngiger Navigationsinstruktionen
• Sprach-Interface für die URWalking-Android-App mit Chatfunktion
• Crowdsourcing von Landmarken
• Bewertung von Landmarken als games with a purpose

Kontaktperson: Prof. Dr. Bernd Ludwig

Titel: CookBERT - Kochen mit BERT

Die Vorstellung des bekannten BERT-Papers von Devlin et al. hat im Jahr 2019 eine kleine Revolution im NLP-Bereich ausgel?st. Denn im Vergleich zu anderen Deep Learning-Modellen kann BERT mit Hilfe seines Self-Attention-Mechanismus Wortbedeutungen und Kontext besser erfassen. Dadurch konnten bis dato aktuelle State-of-the-Art-Modelle, etwa im Bereich des Question Answering und der Sentiment Analyse, bei weitem übertroffen werden.

W?hrend der Fokus zu Beginn haupts?chlich auf englischen BERT-Modellen lag, gibt es diese mittlerweile auch für die franz?sische (CamemBERT) und deutsche (GermanBERT) Sprache. Solche Modelle sind meistens auf der Wikipedia und Nachrichtenartikeln trainiert, d.h. sie bilden das Vokabular und den Kontext in dieser Textform entsprechend gut ab. M?chte man BERT für bestimmte Dom?nen, wie z.B. das Kochen, nutzen, fehlt Information über dom?nenspezifisches Vokabular und deren (kontextueller) Verwendung, was sich m?glicherweise negativ auf die Performance auswirkt.

Im Rahmen einer Abschlussarbeit soll für die Kochdom?ne ein deutsches (und englisches) BERT-Modell erstellt werden. Darüber hinaus soll überprüft werden, ob dieses im Vergleich zu anderen BERT-Modellen zu einer Verbesserung bei Klassifikationsaufgaben im Kochbereich, wie der Vorhersage von Informationsbedürfnissen w?hrend des Kochens, führt oder nicht.

Anforderungen:

• ??? Interesse an NLP- und Machine Learning-Problemen
• ??? Gute Python-Kenntnisse
• ??? Idealerweise bereits erste Erfahrungen mit Keras

?
Kontaktpersonen: Udo Kruschwitz, David Elsweiler und/oder Alexander Frummet.