Online Library

Much research aims to detect online radical content mainly using
radicalisation glossaries, i.e., by looking for terms and expressions associated with
religion, war, offensive language, etc. However, such crude methods are highly
inaccurate towards content that uses radicalisation terminology to simply report on
current events, to share harmless religious rhetoric, or even to counter extremism.
Language is complex and the context in which particular terms are used should not
be disregarded. In this paper, we propose an approach for building a representation
of the semantic context of the terms that are linked to radicalised rhetoric. We
use this approach to analyse over 114K tweets that contain radicalisation-terms
(around 17K posted by pro-ISIS users, and 97k posted by “general” Twitter users).
We report on how the contextual information differs for the same radicalisationterms
in the two datasets, which indicate that contextual semantics can help to
better discriminate radical content from content that only uses radical terminology.
The classifiers we built to test this hypothesis outperform those that disregard
contextual information.

In setting out the work program for the past 12 months, the shared objective of the group was to collaborate across industry, government, and civil society “to map contentsharing algorithms and processes used by industry that may facilitate consumption of content that may increase user interest in or amplify terrorist and violent extremist content and consider positive interventions and risk mitigation points.” This report focuses on the second part of the objective in considering positive interventions and risk mitigation points. GIFCT’s mission is focused on preventing terrorists and violent extremists from exploiting digital platforms. For the purposes of this work we have taken a content-agnostic approach and sought to map positive interventions already used in the countering violent extremism (CVE) space as well as those that have been used elsewhere but could be applied to CVE. The use of counter and alternative narratives is recognized as an online CVE mechanism for connecting users to alternative sources of information. Yet there is less awareness of the uses of other forms of positive interventions.

In setting out the work program for the past 12 months, the group’s shared objective was to collaborate across industry, government, and civil society “to map content-sharing algorithms and processes used by industry that may facilitate consumption of content that may increase user interest in or amplify terrorist and violent extremist content and consider positive interventions and risk mitigation points.” This report focuses on the first part of that objective: To map content-sharing algorithms and processes used by industry. The potential role of content-sharing algorithms in radicalization and violent extremist recruitment continues to be an issue of focus. Yet there is little understanding about what those algorithms are and how they might be misused, contribute to radicalization, or be exploited by violent extremist groups and bad actors. This document is the first step in contributing to that understanding.

The multistakeholder Global Network Initiative (GNI) reviewed more than twenty recent governmental initiatives that claim to address various forms of online harm related to user-generated content — a practice we refer to broadly as “content regulation.” We focused on proposals that could shift existing responsibilities and incentives related to user-generated content. Our analysis illustrates the ways that good governance and human rights principles provide time-tested guidance for
how laws, regulations, and policy actions can be most appropriately and effectively designed and carried out. Because content regulation is primarily focused on and likely to impact digital communication and content, we use international human rights principles related to freedom of expression and privacy as our primary lens.

We welcome the increased focus amongst policymakers on the role played by content personalisation and other algorithmic recommendation systems on online platforms. Such scrutiny is warranted. Terrorist groups exploit platforms that make use of recommendation algorithms, and there are examples of individuals coming into contact with terrorist and violent extremist content via platforms using content personalisation. However, we are concerned that the current debate is, on a policy level, based on an incomplete understanding of terrorist use of the internet, and that a focus on content personalisation is a distraction from more important steps that should be taken to tackle terrorist use of the internet.

Casi 1.5 millón de personas emplean Facebook en todo el mundo. Aunque la mayoría de las usuarios utilizan el sitio con fines positivos, algunos emplean la plataforma con malas intenciones. En este sentido, Facebook ha creado las normas de la comunidad, un conjunto de políticas que detalla qué tipo de contenido es apropiado publicar y qué queda prohibido. Por ejemplo, Facebook prohíbe y elimina contenido que incite al odio, que define como “contenido que ataca directamente a individuos en función de su raza, origen étnico, nacionalidad, religión, orientación sexual, sexo, género, identidad de género, discapacidad grave o enfermedades”. A pesar de que la incitación al odio está prohibida, en ocasiones ciertos usuarios publican contenido desagradable o perturbador que no infringe las políticas de Facebook.

Para contrarrestar ese tipo de contenido desagradable o extremista, Facebook ha declarado públicamente que el contenido compensatorio (y las herramientas que su plataforma ofrece para promoverlo) desempeñan un papel fundamental en la lucha contra dichas malas prácticas. Facebook opina que esta forma de atajar el problema no es sólo resulta potencialmente más eficaz, sino que además es más probable que tenga éxito a largo plazo.