There are several limitations in the use of OSINT for risk analysis, related to both collection and processing/analysis of information.
Collection of information
- Modus operandi employed. There are differences in the digital footprint of different threats, depending on both internal factors, nature of the threat, and external factors, such as legislation and perceived efficiency of law enforcement/security agencies in the digital space (e.g., countries of transit wherein irregular migration is not criminalized may foster more extensive and less confidential online exchanges related to availability of smugglers for example than countries that criminalize this phenomenon and where people are forced to retreat to more protected spaces online and offline). Moreover, there are significant differences between how the same threat manifests itself in a different geographical space: e.g., the type of irregular migration experienced by Estonia is very different from the irregular migration pervasive in the Mediterranean region. These differences, stemming mainly from the modus operandi employed, directly impact the digital footprint. For example, in the case of irregular migration, there are instances where smugglers advertise their services online (e.g. smugglers in Tunisia/Libya assisting people to cross the Mediterranean by boat) and thus by using OSINT, analysts can extract important insight into the magnitude and likelihood of the threat (e.g. departure points, timings, identify of smugglers, routes employed) or instances where the interaction between smugglers and interested parties occurs almost solely in the real world (e.g. word of mouth) or by employing direct encrypted means of communication, often due to security concerns.
- Validity of information. When it comes to social media content it is very difficult to assess the validity of the information provided. For example, there are numerous accounts selling false documents, but in the absence of any other information it is very difficult to assess whether these belong to individuals who are directly involved in this activity and whether the products advertised are real, or simply copies of other accounts.
Processing/analysis of information
- Language. Often the language employed, especially in social media posts is not written correctly, making it very difficult for AI models to correctly extract relevant information. In addition to typos and other types of similar errors, there are also instances of a language written with different characters in an approximation of the traditional ones (e.g., Arabic written with a mix of Latin letters and Arabic numbers, or Cyrillic written with Latin characters). In addition to this, increasingly more, people engaged in illegal activities (e.g. traffickers) employ symbols to replace text (e.g. emojis). Emojis are used either as a direct representation of the activity being offered (e.g. boat to symbolise boat transport offered for irregular migrants on the Mediterranean route) or to covertly refer to a type of criminal activity (e.g. strawberries or cherries to refer to underage individuals advertised for sex work in cases of human trafficking)
- Intelligence-value of online interaction. In addition to content threat indicators, one can also seek to extract ‘behavioural’ indicators, by seeking to understand the extent to which online engagements (e.g., likes and shares of social media posts) relates to the behaviour of individuals in the real world (e.g., the extent to which liking and/or sharing the post of someone advertising human smuggling services across the Mediterranean may indicate an intent to employ that smuggler’s services). For example, several studies have found that increased online activity, often on Facebook and Twitter, has been associated with subsequent increases in protest attendance later. These has been the case of some pro-democratic movements during the Arab spring (Steinert-Threlkeld et al., 2015) or anti-capitalist and economic inequality protests in the United States and Spain (Bastos et al., 2015). This deliverable argues that an even better source of information on this can be found in the field of marketing and more specifically in studies focused on assessing the impact of brands’ social network content on brand awareness and purchase intention (Dabbous & Barakat, 2020). However, the exact quantification of social media engagement in terms of developing risk indicators is still quite problematic.
Resources
Bastos M. T., Mercea D., Charpentier A. (2015). Tents, tweets, and events: The interplay between ongoing protests and social media. Journal of Communication, 65(2), 320–350. DOI: 10.1111/jcom.12145.
Dabbous, A., Barakat Aoun, K, (2020). Bridging the online offline gap: Assessing the impact of brands’ social network content quality on brand awareness and purchase intention. Journal of Retailing and Consumer Services, Volume 53, 2020.
Steinert-Threlkeld, Z.C., Mocanu, D., Vespignani, A. et al. Online social networks and offline protest. EPJ Data Sci. 4, 19 (2015). DOI: 10.1140/epjds/s13688-015-0056-y.
