airi-utcn/press-review

AIRI@UTCN Press Review

Automated press review platform for the Artificial Intelligence Research Institute at the Technical University of Cluj-Napoca (UTCN).

Monitors Romanian press for mentions of the institute, affiliated researchers, and research units via RSS feeds and Google Search.

Quick start

docker compose up --build

• —Press review: http://localhost:8000/
• —Social graph: http://localhost:8000/graph.html

A collection run fires at startup, then every 30 minutes (configurable in data/config.json).

Project layout

press-review/
├── app/           FastAPI service + scheduler + matching logic
├── data/          JSON files — config, articles, graph, people
├── static/        HTML frontend (no build step)
├── tests/         pytest test suite
└── reference/     Original manual press review (docx)

Configuration

RSS sources — data/config.json → sources

{ "name": "My Source", "rss_url": "https://example.com/feed.rss", "homepage": "https://example.com" }

Keywords — data/config.json → keywords

{
  "institute": ["AIRI", "AIRi@UTCN"],
  "university": ["UTCN", "Universitatea Tehnică din Cluj-Napoca"],
  "research_units": ["Trusted AI", "Robotics and IoT"]
}

Changes take effect on the next collection run; no restart needed.

People — data/people.json

Each entry:

{
  "name": "Firstname Lastname",
  "role": "researcher",
  "research_unit": "Trusted AI",
  "profile_url": "https://airi.utcluj.ro/people/slug",
  "aliases": ["Firstname Lastname", "F. Lastname", "LASTNAME, Firstname"]
}

Aliases can be generated automatically:

from app.people_importer import generate_aliases
generate_aliases("Doina Liana Pisla")

How matching works

1. 1.Normalization — text and terms are NFD-decomposed and lower-cased, making matching diacritics-insensitive ("Călin" = "Calin").
2. 2.Word boundaries — terms are matched with \b…\b so "UTCN" does not match "BUTCNA".
3. 3.Relevance rules:
4. 4.Any institute keyword → relevant.
5. 5.Any research_units keyword → relevant.
6. 6.Any university keyword and at least one tracked person → relevant.

matched_keywords and matched_persons on each article in data/articles.json show exactly what triggered inclusion.

Graph

The social graph (/graph.html) shows:

Edges: MENTIONED_IN, PUBLISHED_BY, AFFILIATED_WITH, MATCHED_BY, CO_MENTIONED_WITH.

The date filter bar (Total / Lunar / Săptămânal / Zilnic) limits visible nodes to the selected period.

API

Read

Write

All write operations persist to data/ and take effect immediately without restart.

Running tests

pip install -r requirements.txt pytest
pytest tests/ -v

Collection interval