Classification models and API used in sustainability.discourses project

This repository contains code and trained/ developed models (statistical/ rule-based) to reproduce the classification API used in the sustainability.discourses project running from 2021 to 2025 at the University of Zürich (Digital Society Initiative).

 

Specifically, the API provides endpoints to:

 

- classify German newspaper paragraph for relevance to SDG 11.2, sustainable urban transport, evaluated and developped for the context of Zürich, Switzerland.

- named entity recognition and rule-based entity linking for organizational actors in sustainable urban transport in Zürich

- rule-based topic classification for sustainable urban transport topics.

 

This repository is versioned, documenting the state of the API at a point in time. The latest version may thus not be the relevant version for reproducibility. We attempt to always explicitly link output of sustainability.discourses (such as scientific articles or web applications) making use of the classification API to the API version used to create outputs.

To reproduce and explore the API locally:

 

- download the zenodo_repo.zip folder

- unpack it

- follow the instructions in README.md within the unpacked folder to build and run the Docker container serving the API

 

The API is implemented in Python (3.9) (specifically using spacy and FastAPI. For reproducibility and in production, it is run in a Docker container. Thus, the only hard software requirement is a working version of the Docker CLI. The easiest way to obtain it at the time of writing is probably by installing Docker Desktop.