This project is dedicated to linking street names to the person they are named after. The street information is taken from OpenStreetMap and linked to person data that was generated using Wikidata (as at May 17th 2022).

Our method was tested on streets from the city of Bremen and the state of North Rhine-Westfalia in Germany. It was able to detect 896 new person links in Bremen and 31.363 new links in North Rhine-Westfalia respectively. Overall, we found 183.022 possible street-person links in all of Germany.

In an effort to correctly predict the namesake for a given street, we extracted street data from OpenStreetMap and generated possible candidates for each street. The candidates are taken from an index database that was created by extracting data from a Wikidata dump. For each candidate we further get properties such as occupations, birth place, place of death etc. The candidates are than classified by a machine learning algorithm, picking the candidate with the highest score above a certain threshold as the final result for each street.

The machine learning algorithm was trained on known streets in Wikidata that had the property "named after". On the test data, it achieved a precision of 95,2% with a recall of 90,6%. When applied on streets in OSM, our method was able to detect 896 new person links in Bremen and 31.363 new links in North Rhine-Westfalia respectively. You can take a look at the findings of this project in the following two files:

best_candidates_bremen.csv
best_candidates_nrw.csv

• Python >= 3.8
• Pandas
• Geopandas
• Pickledb
• Scikit-Learn
• qwikidata

To train the ML model and evaluate it, simply run the following command:

ml/ml_training.py

If you want to calculate the results of the approach on OSM, you can generate a file with all candidates for streets in Bremen or North Rhine-Westfalia using the following command. Alternatively, you can use your own file from a different region.

street_candidates.py [ bremen | nrw | file ]

You can then predict the most likely person from the aforementioned candidates for each street.

predict.py [bremen | nrw | file]

The code will return a csv-file containing the candidate with the highest score above a certain threshold for each street.