This project will investigate the role of alternative online subcultures in the promotion of violence, through a data-driven comparative analysis of content across several (Chan) platforms.

Our approach will leverage the outputs of two projects:

• The ESRC CREST project that looks at the use of memes to promote violence [1];
• The EPSRC DADD project that aims at detecting biases in language corpora [2].

First, we will examine the data from 10 different fringe communities and the memetic narratives studied as part of [1]. We will then adapt the framework developed in [2] to systematically analyse the promotion of violence in these communities.

This project will deliver biased concepts that aggregate and capture semantic relations between words with violent connotations towards certain groups, without prior knowledge of hateful words or an early notion of the slang used in a community. Text speech will be analysed in conjunction with visual memes. This will then be used to produce a tool that can tag non-standard harmful content to protect citizens, particularly minors, when browsing the web or chatting online.

We will also investigate the stylometry of posts in each Chan communities. We will use machine learning techniques to trace linguistic traits and measure the type of users in and size of these communities. This will shed light on the demographics of a community that operates in the shadows. Understanding the demographics of these online communities will help in the implementation of educational countermeasures targeted to vulnerable social strata.