KID2 Spot Application

Introduction

We present Spot, a natural language interface for querying OpenStreetMap data and finding Spots, combinations of objects in the public space. Users' natural sentences are translated by a transformer model and turned into OSM database searches. The primary use case is geo-location verification, other use cases however are also possible.

This is a provisionary readme, more information will follow.

How it works

This is a collection of repositories and scripts linked to the KID2 Spot application. Please visit the readme pages of all subtasks and repositories for more details.

Spot Pipeline

The general idea of the pipeline is the following: 1) Bundle similar tags, assign natural language descriptors to create better semantic connections between language and the OSM tagging system 2) Generate random artificial queries including area definition, objects (incl. tags and descriptors) and relations/distances 3) Call GPT API to generate artificial natural sentences from the draft 4) Use Llama3 to extract relevant information from sentences 5) Perform Postgres query in OSM database & display in geographic map

The GUI allows for the display of results of the natural language query in a geographic map, the editing of individual parts of the query, the integration of OpenStreetMap to inspect the candidate locations and more.

Watch the demo video below to get an impression of some of Spot's functionality.

https://github.com/dw-innovation/kid2-spot/assets/23077479/110e3ef0-6fc6-4458-907a-0af5fa377370

Call for participation

This is an open-source project whose code and eventually also website will always be publicly available. If you would like to contribute to the project, please reach out to us via this repository. We would especially appreciate the help of people with in-depth knowledge of the OSM tagging system.

Potential tasks:

Improvement of bundle database, to increase the number of tags included and the quality of the bundles
GPT prompt engineering to write better prompts that result in a more diverse language output
Coding & model development
User testing
etc.

Feel free to reach out if you want to participate in this project.

Contact

If you would like to participate or get in touch with us, please do so by opening an issue, or via email: lynn.khellaf@dw.com

Publications

Proceedings of OSM Science 2023: https://zenodo.org/records/10443346 (also available at https://arxiv.org/abs/2311.08093) Other publication will be shared once available.

Acknowledgments

This project is led by the Deutsche Welle Research and Cooperation Projects teams and was co-funded by BKM (”Beauftragte der Bundesregierung für Kultur und Medien,” the German Government’s Commissioner for Culture and Media).

Map data copyrighted OpenStreetMap contributors and available from https://www.openstreetmap.org.

dw-innovation / kid2-spot