Philipp Dowling

Selected Projects

Screenshot 2019-06-20 at 18.40.48.png

EU Discourse Analysis

I supported Political Scientist Annie Groth on research for her work “Space and the State“, conducting data-driven research into the discourse of the EU Parliament and Commission documents regarding its borders, migration policies, and related issues.

This work represents a practical application of NLP techniques (including Latent Semantic Analysis and custom bayesian models) to a real-world dataset to better inform insights into contemporary political issues and to understand discourse trends on a macro-level.

The paper is available here, the code used to conduct the analysis can be found here.


A browser extension which allows users to pull in contextual information on news stories (related stories, information on people and organizations mentioned in the story) directly in an overlay. Users can also take notes and collaboratively comment on any web content through the extension.

This project showcases the potential that modern semantic technologies offer to enrich web content and to automate what would otherwise be tedious manual research.

This project originally started under the name Convene at HackZurich 2015, where we won the 2nd overall place. Along with our own news crawling and NLP, we integrated DBpedia to source knowledge on different entities in text. The extension itself was built for Chrome using React.

You can find my original pitch from HackZurich 2015 here.




Exploring filter bubbles and automating the discovery of content that allows you to break out of them.

In essence, we attempted to cluster articles on identical or similar events by their relative sentiment polarity to discover articles and sites that exhibited different biases. We presented the result as an interactive reader that would surface alternative sources and reporting on any given topic.

This project was done at hackaTUM in 2016, where it was one of the winners of the wild track. For more information, see the project’s DevPost page.

DBpedia Spotlight Entity Disambiguation

I worked on improving DBpedia’s entity disambiguation engine as part of Google Summer of Code 2015. Entity disambiguation is essentially the task of selecting the correct entity based on the context that a given entity (e.g. person or company) is mentioned in - for example, being able to tell that “Cal beats Washington” refers to the University’s football team, not George Washington or the state of Washington.

I implemented an approach based on a log-linear learning-to-rank model using a combination of semantic and statistical features, leading to an increased overall MRR performance. As part of this project, I also improved the entity vectors generated using wiki2vec.

This project was a great learning opportunity for remote open source collaboration, as well as diving deeply into knowledge graphs and entity linking technology.

Progress repo here. DBpedia codebase is here.