This is the repository of my app, which you can visit at: https://nyhedsoverblik.streamlitapp.com/
It is a news app, that every day at 9AM fetches the headlines from Danish news media: DR, TV2, Jyllandsposten, Berlingske, Politiken, and Ekstra-Bladet. Based on the headlines, I run a clustering analysis and estimate today's topics.
The app is hosted on Streamlit cloud. Data is fetched using a scheduled Github Action. The clustering analysis is done using DBSCAN on a similarity matrix (Word Mover's Distance) based on Skip-Gram embeddings pre-trained on the CoNLL2017 dataset (thanks DaNLP!). The description of each topic is made using a pre-trained summarization model.