Internet-based communication systems have become an increasing tool for spreading misinformation and propaganda. Even though there exist mechanisms that are able to track unwarranted information and messages, users made up different ways to avoid their scrutiny and detection. An example is represented by the mixed-code language, that is text written in an unconventional form by combining different languages, symbols, scripts and shapes. It aims to make more difficult the detection of specific content, due to its custom and ever changing appearance, by using special characters to substitute for alphabet letters. Indeed, such substitute combinations of symbols, which tries to resemble the shape of the intended alphabet’s letter, makes it still intuitively readable to humans, however nonsensical to machines. In this context, the paper explores the possibility of identifying propaganda in such mixed-code texts over the Internet, centred on a machine learning based approach. In particular, an algorithm in combination with a deep learning models for character identification is proposed in order to detect and analyse whether an element contains propaganda related content. The overall approach is presented, the results gathered from its experimentation are discussed and the achieved performances are compared with the related works.

Art.No.: 2196; This paper is an extended version of the paper published in the 15th International Conference on Availability, Reliability and Security (ARES 2020), Virtual Event, Dublin, Ireland, 25–28 August 2020; Article 76