Diego Gomes' MSc thesis received a distinction from the MNF
Diego Gomes recently completed his Master thesis entitled "Geoparser: A Transformer-Based Bi-Encoder Approach for Efficient Toponym Disambiguation". In his thesis, he developed algorithms which identify and assign coordinates to placenames found in text. So, given an input like "The cyclists started their ride from Bellevue, and finished at Helvetiaplatz" the tool should find two placenames (Bellevue and Helvetiaplatz) and give them coordinates, such that they can be placed on a map.
For a human, reading this text, and living in Zurich, this is an easy task. We all know where Bellevue is, and most of us are familiar with Helvetiaplatz. Thus, even though the text doesn't mention Zurich, we are pretty sure about the places being referred to. This is much harder for a computer - there are many Bellevues not just in Switzerland, but all over the world, and its not so surprising that there is also, for example, a Helvetiaplatz in Luzern. To solve this problem Diego used machine learning based techniques which try to use contextual clues, just like humans. The approach he developed not only worked, but was computationally more efficient than many other current techniques.
Diego's work was judged by the MNF to be worthy of a distinction - a prize awarded at his graduation this spring. This doesn't happen very often, and we are proud that Diego's achievements have been recognised. We're equally proud though that Diego published his code as an Open Source library in Python (https://geoparser.app). The code has already been used by other researchers in their work, and as such Diego's work is having real impact in our field.