Studentenvorträge zu Projekt- und Abschlussarbeiten:

 

Raul Pinto: Machine learning user preferences in tag-enabled social networks

Diplomarbeit - betreut von Gefei Zhang

 

Data mining algorithms try to discover knowledge in data. There are several algorithms available with a diverse set of applicability and performance. Many algorithms are combined into libraries to easily test them with different data. Those libraries may be included into new applications. aiD Me Tweet is an application, that incorporates one of these libraries, WEKA, to test these algorithms on their applicability to small texts. With aiD Me Tweet, it is not only possible to gather real-life data from Twitter, but also to test generated schemes on their performance. The intention of this thesis is to evaluate currently available data mining algorithms by their applicability to Tweets. For this, we introduce Twitter and knowledge discovery in databases as well as Grails, a framework for web applications. Furthermore we share our experiences about working with the Twitter API and WEKA. We include an application for evaluating the performances of different data mining algorithms to short texts. During development, we extended our goals and also lowered the time needed, until a user profits from the discovered knowledge. We discuss different approaches on how to increase performances before we conclude with our results and possible future works.

 

Xiaomin Song: Aspekt-orientierte Modellierung von Rich Internet Applications

Diplomarbeit - betreut von Gefei Zhang

 

Seit einiger Zeit erleben wir einen großen Boom der sogenannten Rich-Internet-Applications (RIAs). Die heutigen Web-Anwendungen zeigen nicht mehr nur statische Informationen an, sondern bieten auch elaboriertere Features wie zum Beispiel interaktive UI oder asynchrone Client-Server-Kommunikationen an. Dies stellt die Modellierungsansätze vor neue Herausforderungen, da die Kommunikationen zwischen unterschiedlichen Widgets sogenannte Cross-cutting-Features sind und schwer separat modelliert werden können. In dieser Arbeit entwickeln wir einen aspekt-orientierten Modellierungsansatz für RIAs, der es erlaubt, verschiedene logische Aspekte eines Widgets getrennt zu entwerfen und zu entwickeln. Bei der Modellierung werden im ersten Schritt die Widgets separat modelliert, ohne das Zusammenspiel zwischen den Widgets zu berücksichtigen. Die Widgets werden durch UML-Zustandsmaschinen dargestellt. Im zweiten Schritt werden die Aspekten mit der im Rahmen dieser Arbeit definierten AspektScript-Sprache formulieret, um das Zusammenspiel von den Widgets zu modellieren. Die Aspekte werden durch den (automatischen) Weavings-Prozess zu unterschiedlichen Zustandsmaschine-Elementen übersetzt und mit dem Zustandsdiagramm zusammengefügt. Eine hoch-modulare Modellierung von RIAs wird daher durch unseren Ansatz ermöglicht. Die praktische Anwendbarkeit unseres Ansatzes wird in einer Fallstudie validiert.

 

Thomas Mair: A Framework for Evaluating and Selecting Actions in Pervasive Adaptive Applications

Masterarbeit - betreut von Andreas Schröder und Christian Kroiß

 

In the last ten years pervasive computing has had a tremendous impact on many people's daily life. More and more people have access to the internet through pervasive devices. Despite the progress in pervasive computing research and technology the communication between the user and the device only slightly changed since the early days of personal computing. The affective computing research tries to improve the communication between humans and computers by providing the computer not only with the user's input but also with its cognitive and emotional state. This thesis presents a framework for applications that process information about the user's physiological state. The emphasis of the framework lies in evaluating a set of actions by considering the physiological state of the user and in the recording of effects caused by selecting a certain action. The main design goals are easy addition of new algorithms, generality and simplicity. In addition to the main goal of the framework also basic tooling support for developing affective software with the framework is being provided. To evaluate and test the design of the framework three algorithms adopted form artificial intelligence and data mining have been implemented. An example evaluation has been performed to study the tooling support and to provide basic ideas on how algorithms in an affective application may be analyzed. The framework itself is built on top of the REFLECT middleware, which provides means to record and analyze data from sensors in a component based architecture.