Build a network of co-occurring cities in publication affiliations. Visualise the network and find interesting relationships. Which are the strongest co-occurrences across countries? Does geography play a role? Use relative optionally log-odds ratios in your analysis. Draw the network on a map.
python3 -m venv .venv # create virtual environment
source .venv/bin/activate # activate virtual environment
pip install -r requirements.txt # install dependencies
python data_collection.py # to fetch some papers and coordinates
python main.py # to build graph, run visualization, ...- papers and coordinates fetched from api can be saved locally with methods in
file_io.py(files can be found indata/papersanddata/coordinates) - a list with countries names, coordinates, codes and alternative names can be found in
data/dictionaries
- How should we use log odds ratios? What is meant by "relative optionally"?
- Concerning "interesting relationships" in the assignment: One idea we had was finding out which countries have a particularly well research in a specific topic. Are we on the right track? Probably that would be a use case for log odds ratios?
- Is it crucial to have great performance when analyzing the data? How big should the dataset be?
- How much accuracy is expected when extracting city names? Probably there will always be affiliations where we won't be able to extract a city and country. (Affiliations don't have a uniform format.)
- (How to fetch all papers from PubMed at once?)
- improve city visualization (Florian)
- visualize relationships between countries (Florian)
- improve location extraction (Oliver)
- make coords fetching faster (persist city coordinates) (Albrecht)
- try to find interesting relationships (e. g. per research field) (Markus)
- special tasks for Karl (Karl)