Topics for Projects and Thesis @IWM/KC

Linked Open Data (LOD) is an initative to interlink data sources available in the web by using semantic web technologies, in particular RDF. It targets to establish a Web of data, similar to the Web of documents today.
In their work, students can explore the following topics

1. Develop a application scenario using LOD data and implement a web application therefore
2. Develop a web service & web application to semi-automatically create links between different LOD repositories
3. Develop and analyse ontology alignment algorithms for automaticall creating links between different LOD repositories
4. Discovery and add new data sets and interlink them with the LOD cloud (see for example RIESE)
5. Analyse the quality of LOD repositories and their interlinkage

Relevant Literature:

Bizer C, Heath T, Berners-Lee T. Linked data–the story so far. International Journal on Semantic Web and Information Systems. 2009. Available at: http://tomheath.com/papers/bizer-heath-berners-lee-ijswis-linked-data.pdf

Shadbolt N, Berners-Lee T, Hall W. The Semantic Web Revisited. IEEE Intelligent Systems. 2006;21(3):96-101. Available at: http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=1637364.

With the growing amount of information its quality decreases. Furthermore, the populartiy of social networks offers a broad field for misuse, ranging form simple spam to user masquarades. Within this topic, students should explore either plagiarism analysis or misuse in social networks. Work covers the development of services countering misuse or plagiarism detection or the analysis of respective data set and new algorithms.

Details, data sets and algorithms on the topic can be found on the PAN09 Workshop site.

Contact: Karin Schöfegger (k.schoefegger@tugraz.at)

Als Folksonomy (oder collaborative tagging) wird die gemeinsame Verschlagwortung von diversen Resourcen bezeichnet. Bekannte Beispiele im Web sind del.icio.us (für Bookmarks), flickr (für Fotos), youtube (für Videos) etc. Im Laufe der Zeit verändert sich die entstehende Struktur aus Tags, Usern und Resourcen des zugrundeliegenden tripartiten Graphen. Neue User kommen zur Community hinzu, neue Resourcen werden annotiert, nutzt ein User ein Tagging-System länger, ändert sich auch sein Tagging-Verhalten.

Ziel der Arbeit ist es, die zeitliche Analyse von Folkonomies zu unterstützen. Aus ausgewählten Datenquellen sollen zu bestimmten Themenkreisen regelmäßig aktuelle Daten bestehend aus Usern, Tags, Resourcen und deren Verbindungen eingelesen und (als Graphenstruktur, zB GraphML) gespeichert werden. Diese Graphenstruktur soll in Abhängigkeit des gewählten Zeitpunkts dargestellt werden können. Zusätzlich sollen je nach Interesse versch. Algorithmen implementiert und ausprobiert werden die diese Struktur zu einem bestimmten Zeitpunkt analysieren, zB Clustering Algorithmen, Similarity Measures, etc. oder untersucht werden, wie sich beispielsweise Cluster zeitlich verändern.

Programmiersprachen: Python (zB igraph) oder Java (zB. Jung)

Klassische digitale Enzyklopädien ermöglichen Anwendern die stichwortbasierte Suche nach Artikeln und die Navigation zwischen Artikeln mittels Querverweisen. Moderne multimediale Enzyklopädien sind zunehmend semantisch strukturiert und erfordern daher erweiterte Such- und Navigationsmöglichkeiten. Die digitale Enzyklopädie des Brockhaus bietet eine dreidimensionales System zur Navigation zwischen Artikeln. Der OCZ Neural Impulse Actuator (http://www.ocztechnology.com/products/ocz_peripherals/nia-neural_impulse_actuator) ermöglicht kostengünstig die Nutzung von Gesichtsmuskelspannung, Augenbewegung und Gehirnströmen als Eingabemedium für Standard-PCs.
Inhalt der Arbeit ist der Entwurf von adaptierten Interaktionsformen für den 3D-Wissensraum der Brockhaus Enzyklopädie, welche die Verwendung des OCZ Neural Impulse Actuator für Anwender ermöglichen sowie die Evaluierung der entwickelten Interaktionsformen im Vergleich zu Standard-Eingabegeräten (Maus, Tastatur).

Klassische digitale Enzyklopädien ermöglichen Anwendern die stichwortbasierte Suche nach Artikeln und die Navigation zwischen Artikeln mittels Querverweisen. Moderne multimediale Enzyklopädien sind zunehmend semantisch strukturiert und erfordern daher erweiterte Such- und Navigationsmöglichkeiten. Die digitale Enzyklopädie des Brockhaus bietet zwei visuelle Navigationssysteme, die sich in der Dimensionalität der Darstellung und in den verfügbaren Interaktionsformen unterscheiden.
Inhalt der Arbeit ist die Planung, Durchführung und Auswertung einer Benutzerstudie, welche - aufbauend auf bereits gewonnenen Ergebnissen und erprobten Vorgehensweisen - die Vor- und Nachteile beider Systeme beleuchtet.

Contact: Andreas S. Rath (arath@know-center.at), Didier Devaurs (ddevaurs@know-center.at)

The topic “Online Task Detection” belongs to the research field of User Context Detection. The term “user context” is used to describe a user’s computational environment (e.g. user interactions, current task, recently opened documents, work environment). User context is described in formal “user context models” and may be utilized for task recognition, context-based information retrieval or context-aware applications.
For a Bachelor or Master Thesis or Project in the field of User Context Detection, you need basic programming skills and interest for the following fields

For a Bachelor or Master Thesis or Project in the field of User Context Detection, you need basic programming skills and interest for the following fields

Human Computer Interaction, Personal Information Management, Semantic Technologies, Machine Learning and Context-Aware Information Retrieval

Short description of the project “Online Task Detection”

• A task is a piece of work that needs to be done by a user. User context observation mechanism can capture those parts of the task execution that involve interactions with the user’s computer desktop, applications and resources. Based on these observations task detection should detect which task is currently performed by the user.
• Detecting the task a user is performing at the moment is challenging because the user can utilize multiple resources or can switch between various tasks (multi-tasking).
• The goal of the project is to find and implement algorithms that can detect which parts of the observed user context belong to the same task. The evaluation of the algorithms can be done on existing user context datasets and should be implemented into an already existing prototype that is based on Java and Eclipse RCP.

For an overview of all topics related to User Context Detection see User Context Detection - Themen SS 2009

Contact:  Didier Devaurs (ddevaurs@know-center.at), Andreas S. Rath (arath@know-center.at)

The topic “User-oriented Task Detection” belongs to the research field of User Context Detection. The term “user context” is used to describe a user’s computational environment (e.g. user interactions, current task, recently opened documents, work environment). User context is described in formal “user context models” and may be utilized for task recognition, context-based information retrieval or context-aware applications.
For a Bachelor or Master Thesis or Project in the field of User Context Detection, you need basic programming skills and interest for the following fields

For a Bachelor or Master Thesis or Project in the field of User Context Detection, you need basic programming skills and interest for the following fields

Human Computer Interaction, Personal Information Management, Semantic Technologies, Machine Learning and Context-Aware Information Retrieval

Short description of the project “User-oriented Task Detection”

• A task is a piece of work that needs to be done by a user. User context observation mechanism can capture those parts of the task execution that involve interactions with the user’s computer desktop, applications and resources. Based on these observations task detection should detect which task is currently performed by the user.
• The notification of a user that a new task has been detected may interrupt the user’s work, such that is important to reduce false and too often notification.
• The goal of the project is to study “when” and “how” a task should be detected and presented to the user for confirming or refining the detection result. The gathered insight should be implemented into an already existing prototype that is based on Java and Eclipse RCP.

For an overview of all topics related to User Context Detection see User Context Detection - Themen SS 2009

Contact: Didier Devaurs (ddevaurs@know-center.at), Andreas S. Rath (arath@know-center.at)

The topic “Context Detection in Web Browsers” belongs to the research field of User Context Detection. The term “user context” is used to describe a user’s computational environment (e.g. user interactions, current task, recently opened documents, work environment). User context is described in formal “user context models” and may be utilized for task recognition, context-based information retrieval or context-aware applications.
For a Bachelor or Master Thesis or Project in the field of User Context Detection, you need basic programming skills and interest for the following fields

For a Bachelor or Master Thesis or Project in the field of User Context Detection, you need basic programming skills and interest for the following fields

Human Computer Interaction, Personal Information Management, Semantic Technologies, Machine Learning and Context-Aware Information Retrieval

Short description of the project “Context Detection in Web Browsers”

• The web browser has become a key application on nowadays desktop computers which is used for various types of tasks. Hence it would be interesting to observe how users utilize the browser and how they interact with it when they perform their tasks.
• Since it is tedious and time consuming for the users to manually have enter and describe what they are doing it is a necessity to develop mechanisms that are able to automatically collect context data about the users.
• The goal of this project is analyze what kind of user context information can automatically be collected in a web browser and to develop plug-ins for web browsers to do so. User context data can be but is not limited to the currently viewed web page, mouse movements, text and mouse inputs, text highlighting, text modifications, page content that is currently in focus, bookmarking, reloading a webpage, forward and backward navigation, printing, using the search box or copy & cut & paste interactions.

For an overview of all topics related to User Context Detection see User Context Detection - Themen SS 2009

The better the intention behind a query can be understood, the better the corresponding information need can be satisfied - this hypothesis appears to be reasonable and a lot work of is dedicated to studying the “Search Intent”. Different notions of search intent have been introduced and corresponding approaches to classify queries according these notions have proven to be successful. Future directions to explore include finding and providing appropriate actions to be taken once the intention behind a query is approximated. You will devise and implement an algorithm that suggests potential next steps (means) to be taken in order to fulfil the searcher’s intention.

The question of how to represent and rank Web results has been addressed in different ways. Recently, Google has launched an application that allows a personal re-ranking of Google results thereby personalizing parts of your search. You will implement algorithms to cluster web search results that will be integrated into an existing framework already providing web search functionality. A further objective is to examine Web search results from an intentional perspective, e.g. can they be clustered or re-ranked by taking into account the presence of intentional patters. Techniques such as “Probabilistic Latent Semantic Indexing (PLSA)” will be explored to evaluate query - cluster dependencies. Related Literature: [1]