1. Predicting Merger Decision Outcomes of the European Commission: A Natural Language Processing and Machine Learning approach
Master Thesis, Hertie School (October 2022 — May 2023)
Github repo: predicting-merger-decision-outcomes
A novel application of Natural Language Processing and Machine Learning to predict merger decision outcomes of the European Commission by analyzing extracted text from official merger decisions. Identified a Support Vector Machine linear classifier as the best-performing model in predicting mergers that were either approved unconditionally or subject to conditions with an 84% recall and 63% precision. Read more on this post.
2. Analyzing the German Public Procurement Network: Implications for Corruption Risk
A Research Project for Applied Network Analysis, Hertie School (April 2023 — May 2023)
Github repo: network-analysis-EU-procurements
Analysis of corruption risk using single-bidding rates in centralized German public contracts 2008-2016 by conducting network-based approaches such as R-A clustering, degree distributions, k-core decomposition, modularity maximization, and measurement of nodal centrality. Read more on this post.
3. Classification of Unstructured Documents: Transfer Learning with BERT
A Tutorial Project for Deep Learning, Hertie School (November 2022 — December 2022)
Github repo: EU-DMA-Text-Classification-BERT
A Python tutorial for building a text classification pipeline using unstructured documents with Bidirectional Encoder Representations from Transformers (BERT). The tutorial focuses on predicting stakeholder agreement with the EU’s Digital Markets Act proposal using public consultation documents.
4. Emotionality and Rationality in Speeches of EU Commissioners
A Research Project for Text as Data , Hertie School (November 2022)
Github repo: Emotionality-in-Political-Speech-of-the-EU-Commission
Investigated the variation in the use of reason and emotion in communicating politics of the European Commission across time and events using word2vec word embeddings and structural topic modeling.
5. High Resolution Traffic Accident Prediction in Berlin
A Research Project for Machine Learning, Hertie School (February 2022 — May 2022)
Github repo: Traffic-Accident-Prediction-Berlin
Predicted the occurrence of accidents on a given road segment at a given date and hour in Berlin by training machine learning models such as Logistic Regression, Random Forest, Balanced Random Forest, Balanced Bagging, Support Vector Classifier, and XGBoost using data on road accident 2018-2020, road network, and historic weather.
6. Spotify-MatchedR
A Research Project for Machine Learning, Hertie School (November 2021 — December 2021)
Github repo: Spotify-MatchedR
An R shiny app for measuring music taste compatibility of 5 users based on their 100 most listened songs 2018-2020 using Spotify’s Web API. Musical metrics like danceability and popularity were visualized using plotly. Try out the app here: Spotify-MatchedR.
7. GetHelp
A Project for Data Structures and Algorithms, Hertie School (September 2021 — December 2021)
Github repo: GetHelp
A Python-based app built using object-oriented programming to match requests (e.g., ride, tutor) of students of Hertie School.