Here's a list of projects I have been working on. Some are university projects, some are just smaller tools I have developed for my own needs. You can find more things on my GitHub account.
Master's Thesis 2014
My master's thesis:
"Weakly Supervised Learning for Compositional Sentiment Recognition"
My master's thesis was supervised by Dr. Yannick Versley. Prof. Dr. Anette Frank was the second examiner. Please see the separate page on my ma-thesis for more information.
Bachelor's Thesis 2011
My bachelor's thesis:
"Analyse von Netzwerken zwischen Pharma-Firmen sowie von klinischen Studien auf die Frage, ob befreundete Firmen noch ihre Produkte gegeneinander testen"
"Analyzing a pharma company network: Are friends still competing?"
My Bachelor's thesis was supervised by Prof. Dr. Stefan Riezler and Prof. Dr Gerhard Reinelt. Research was done in the junior research group "Network Analysis and Graph Theory" led by Dr. Nina Zweig at the Interdisciplinary Center for Scientific Computing.
Most of these projects were done as part of course work and are not necessarily production quality. They are intended to serve as proof of concept for design and algorithm implementation.
Multimodal Distributional Semantics 2013
Traditional distributional models of meaning are built from large text corpora. However, many things go unmentioned in text. As McClelland puts it, the traditional approach to distribution semantics is like "learning meaning by listening to the radio". By integrating other modalities, such as images, we can improve our semantic models.
In this project, I try to enhance prediction of semantic relations between head nouns and modifier nouns by integrating pictoral data from ImageNet. For example, the QUALITY relation holds for the noun phrase "brick house". I extract image features from pictures showing bricks and houses to improve prediction performance. Image features are extracted as SIFT descriptors using the OpenCV framework. The code is written in Python and C++.
This project was done as course work for the course "Distributionelle Semantik jenseits der Wortbedeutung". The multimodal approach is inspired by presentations from the seminar "Multimodale Semantik".
Twitter Sentiment Analysis 2013
Social Media Monitoring has become an important research focus. Applications range from reputation monitoring to earthquake detection.
In this project, we adapt and extend existing approaches to the German language to build a sentiment polarity classifier for German tweets. As a basic use case, we would like to get a feeling if the sentiments towards a currently trending topic such as '#tatort' are positive or negative.
This project was done by Tilman Wittl and myself as course work for the course "Text Mining".
Distributed Web Crawler 2012
A distributed web crawler and indexer written in Hadoop. It starts with a small list of URLs and iteratively expands its index from there. A simple boolean query model is supported.
This project was done as course work for the course "Advanced Programming".
GIVE-2: Natural Language Generation in Virtual Environments 2010
The goal of the GIVE-2 challenge is to generate directions in a virtual 3D environment which direct the user to a well-hidden trophy. GIVE-2 systems create directions in natural language based on an abstract plan provided by the GIVE framework.
This project was jointly done by Eric Hildebrand, Eleftherios Matios and myself as course work for the course "Natural Language Generation for Virtual Environments".
RECIPE: Recipe Event Chain Imperative Processing Engine 2010
Narratives contain chains of events. The RECIPE project extracts common event chains from cooking recipes from a basic logic representation of the instructions contained therein.
This project was jointly done by Hiko Schamoni, Tilman Wittl, Britta Zeller and myself as "Softwareprojekt" as required by the Bachelor of Arts curriculum at the University of Heidelberg.
Leechi is a small Python library to automate downloads from web servers without drawing too much attention to oneself.
Leechi was developed as part of my work at the Forschungsdaten Service Center.
Screen Scraping & BeautifulSoup 2013
I have given a presentation on Information Extraction using HTML scraping with Python and BeautifulSoup as part of my work at the Forschungsdaten Service Center. Slides and code are available.