Our process is three-fold, consisting of information gathering, verification, and publication.

Information gathering:

The information gathering is performed systematically through diverse but complementary sources including: 

• Established media networks such as journalists’ associations, unions, reporters in several media outlets, etc; 
• A mobile-friendly ReportIt form which enables anyone to submit alerts to the monitoring team
• The News & Tweets Observer, an Artificial Intelligence tool that automatically detects violations which are reported on Twitter or in news articles (more information below)

Verification

Before being published, the alert is verified by our network of experts. Sources are checked and further information is collected from other platforms and publications and through direct communication. 

Publication

If the incident was successfully verified, it is published as an alert on our platform. The alert contains a description of the incident, is geo-tagged, and classified based on our comprehensive category system, and is then viewable through the Alert Explorer.

Types of incidents: We cover incidents regarding physical attacks, harassment/psychological abuse, attacks to property, censorship, and legal incidents. For more details please see the classification system below.

• • Who:  We focus on threats to journalists and media workers. However, due to the complex nature of threats to media freedom we go beyond this and also document threats addressed to:
    • Persons who are being attacked due to their personal or professional connection to journalists and media workers, such as family members or journalists’ sources;
    • Media companies like media owners, media houses, or publishers;
    • Associations like trade unions or Non-Governmental Organisations fighting for press freedom
    • Threats or attacks aimed at the media in general, such as laws restricting press freedom or blocking access to websites or sources of information.
  • Types of journalists or media actors
    • • We categorise journalists and media actors as follows: journalists/editor, photographer/camera operator, translator/fixer, blogger/citizen journalist, whistleblower, media owner/broadcaster
  • Covered region/countries: Our main focus is incidents happening in European Union Member States and candidate countries. This also includes incidents that happened in these countries against individuals from outside Europe i.e. a non-European journalist who is attacked in a European country. Additionally, MapMF holds some historical data from before the start of MFRR in March 2020, including further neighbouring countries which are not maintained or updated anymore. Following a major category revision performed at the same time in order to reflect all threats faced by media workers, alerts recorded before 2020 may not be comparable to the current alerts. 

Each alert is categorised according to different aspects of our comprehensive category system. This allows for a flexible filtering on the Alert Explorer as well as detailed analysis and statistics which can be provided on the press freedom situation regarding different aspects.

The main aspects we classify:

• Type of attack
• Source of attack (type of aggressor)
• Context of attack (place where the attack happened)
• Who was attacked (such as journalist, media company, family member, journalists’ sources) 

For journalists and media workers, we additionally classify further profile information of the victim if it is available:

• Type of journalist or media actor
• Gender
• Employment status

Furthermore, there are also a number of specific topic labels, which allow labelling an alert when it is linked to this topic, such as::

• Coronavirus/COVID-19/SARS-CoV-2
• SLAPP (Strategic Lawsuits Against Public Participation)
• Environmental reporting
• Racism/Xenophobia
• Gender-based attacks
• Sexual orientation

The category system is regularly updated to reflect all threats faced by media workers. The detailed categories for each aspect above can be viewed within the Alert Explorer by expanding the filter.

As press and media freedom violations are frequently reported in news articles and tweets, these channels provide a valuable source in terms of gaining knowledge of press and media freedom violations. However, a comprehensive monitoring of news and Tweets across Europe is not feasible in a manual process. 

Therefore, we developed the News & Tweets Observer (NTO), an innovative tool trained to detect press and media freedom violations reported within news articles or tweets based on Artificial Intelligence. It was developed by InfAI and ECPMF as part of the MFRR project and is one of our three main pillars for information gathering (see process).

The News & Tweets Observer is currently focusing on text in the English language, providing wide coverage of incidents within our mandate region but being reported worldwide. The first experiments to extend the NTO with further languages led to promising results.

We published two peer-reviewed scientific papers about the News & Tweets Observer, which were presented at international conferences.: 

Tariq Yousef, Antje Schlaf, Benjamin Bock, Janos Borst, Andreas Niekler, Gerhard Heyer: Towards Automatic Detection of Reported Press and Media Freedom Violations in Twitter and News Articles. In: Proceedings of the 8th Conference on Computer-Mediated Communication (CMC) and Social Media Corpora (CMC-Corpora 2021), 2021 

• describes the first part of the development, namely the data gathering, the creation of training data for the topic, and the evaluation of different classification models, including deep learning algorithms with promising results

Tariq Yousef, Antje Schlaf, Janos Borst, Andreas Niekler, Gerhard Heyer: Press Freedom Monitor: Detection of Reported Press and Media Freedom Violations in Twitter and News Articles. In: Empirical Methods in Natural Language Processing (EMNLP), System Demonstrations, 2021  

• describes the second part of the development, namely the developed final tool, its architecture including the trained deep learning model and other components, the user interface, as well as the system evaluation and a usability study
• the virtual presentation of the scientific paper and the software for this conference can be seen here