Several national and international institutions and organizations are committed to monitoring health news in order to derive Public Health Intelligence (PHI). Their primary objective, among others, is to identify emerging health threats. They achieve this by utilizing aggregator platforms such as the Epidemic Intelligence from Open Sources (EIOS) platform. However, their efforts are significantly hindered by the overwhelming amount of textual data, as they need to sift through more than 100 000 multilingual articles daily. To better tailor articles to different user interests (such as source and topics covered), we developed NARPHI, a recommendation system based on Graph Neural Networks (GNN). This system uses user and article features and past interactions (i.e., combined into ratings), recommending the most relevant articles for each user. Our initial findings with NARPHI show that the recommendations using the best combination of user community and article event type features align with user preferences.
FRANCISCO DE SOUSA Diana;
SPAGNOLO Luigi;
STEFANOVITCH Nicolas;
2026-06-30
CEUR-WS.ORG
JRC142027
1613-0073 (online),
https://ceur-ws.org/Vol-4206/INRA-1.pdf,
https://ceur-ws.org/Vol-4206/,
https://publications.jrc.ec.europa.eu/repository/handle/JRC142027,
| Name | Country | City | Type |
|---|
This document is only visible at the Commission level.
You are not authorized to publish or distribute it outside the European Commission.
This is a public document. You can share this publication.
Datasets
| ID | Title | Public URL |
|---|
Dataset collections
| ID | Acronym | Title | Public URL |
|---|
Scripts / source codes
| Description | Public URL |
|---|
Additional supporting files
| File name | Description | File type |
|---|