Periodic Reporting for period 5 - SOCSEMICS (Socio-Semantic Bubbles of Internet Communities)
Reporting period: 2022-09-01 to 2023-08-31
SOCSEMICS aims at developing a set of integrated, groundbreaking methods to address the possible existence of so-called "echo chambers" or "bubbles" in the digital public space. In all generality, this phenomenon may be construed as communities whose actors are more or less prone to interaction and open to informations coming from "outside" of these communities.
The last decade has witnessed a massive growth in the number of studies devoted to the understanding of the informational and interactional structure of web communities, not least for their increasing contribution to the shape of the epistemic landscape of our connected societies — for one, the digital public space has assuredly become a significant part of the public space as a whole: it hosts information dynamics and knowledge construction processes key to contemporary political dynamics and democratic deliberation processes. Debates regarding the specificities of the digital public space, especially with respect to its offline counterpart, have been striving: be it between scholars arguing whether the Internet is a haven for heterodox positions absent from the offline public space or academics discussing whether it reproduces classical power/authority positions and relations and, more broadly, to what extent the digitization of political discussion may facilitate or hamper democratic deliberation. In the meantime, the issue of a possible “balkanization” of digital spaces or the emergence of so-called “echo chambers” has continuously received a significant attention.
SOCSEMICS will contribute to the formalization and operationalization of this question by addressing four current challenges:
• developing a comprehensive theory of reinforcing and self-sustaining confined clusters by appraising the social, semantic and socio-semantic realms simultaneously;
• drastically improving content analysis by replacing classical distributional approach with clause analysis, thus enabling quantitative approaches rendering the linguistic complexity of utterances in web corpuses;
• fostering the interface between these methods and qualitative approaches, especially through a couple of broad case studies, namely on sanitary crises or elections in several European countries;
• interactive platforms implementing the above innovations and facilitating digital social research.
The last decade has witnessed a massive growth in the number of studies devoted to the understanding of the informational and interactional structure of web communities, not least for their increasing contribution to the shape of the epistemic landscape of our connected societies — for one, the digital public space has assuredly become a significant part of the public space as a whole: it hosts information dynamics and knowledge construction processes key to contemporary political dynamics and democratic deliberation processes. Debates regarding the specificities of the digital public space, especially with respect to its offline counterpart, have been striving: be it between scholars arguing whether the Internet is a haven for heterodox positions absent from the offline public space or academics discussing whether it reproduces classical power/authority positions and relations and, more broadly, to what extent the digitization of political discussion may facilitate or hamper democratic deliberation. In the meantime, the issue of a possible “balkanization” of digital spaces or the emergence of so-called “echo chambers” has continuously received a significant attention.
SOCSEMICS will contribute to the formalization and operationalization of this question by addressing four current challenges:
• developing a comprehensive theory of reinforcing and self-sustaining confined clusters by appraising the social, semantic and socio-semantic realms simultaneously;
• drastically improving content analysis by replacing classical distributional approach with clause analysis, thus enabling quantitative approaches rendering the linguistic complexity of utterances in web corpuses;
• fostering the interface between these methods and qualitative approaches, especially through a couple of broad case studies, namely on sanitary crises or elections in several European countries;
• interactive platforms implementing the above innovations and facilitating digital social research.
The notion of socio-semantic cluster (rather than the more general terms “bubble” or “echo chamber”) has been specified in terms of alignment between social and semantic clusters, which could be denoted as socio-semantic cohesion — to what extent users are jointly confined in interactional terms (cohesive groups where interactions occur significantly more between group members) and in informational terms (homogeneous distribution of opinions).
So far, SOCSEMICS could describe the variety of the configuration of some digital public spaces in terms of such clusters, at the level of the whole space (existence of various clusters, interconnection, dispersion and fragmentation) and at the user-level (in particular, how users variously concentrate their attention on a variety of other users and topics). For instance, several case studies could be defined, pertaining to the European Elections in 2019 or debates around the sanitary crisis in 2020, in English-, French-, German- and Italian-speaking portions of Twitter.
In parallel, SOCSEMICS could make significant progress on text processing in order to ultimately enrich the socio-semantic ontology with the attribution of opinions or beliefs to actors. Put simply, it makes it possible to deal with “subject-verb-object” stuctures, and beyond, at any arbitrary level of complexity, rather than word bags or word distributions which leave a large part of the interpretation to the observer. SOCSEMICS introduced the notion of “semantic hypergraphs” which enable sophisticated computations on claims. An open-source platform called “graphbrain” implements these concepts.
So far, SOCSEMICS could describe the variety of the configuration of some digital public spaces in terms of such clusters, at the level of the whole space (existence of various clusters, interconnection, dispersion and fragmentation) and at the user-level (in particular, how users variously concentrate their attention on a variety of other users and topics). For instance, several case studies could be defined, pertaining to the European Elections in 2019 or debates around the sanitary crisis in 2020, in English-, French-, German- and Italian-speaking portions of Twitter.
In parallel, SOCSEMICS could make significant progress on text processing in order to ultimately enrich the socio-semantic ontology with the attribution of opinions or beliefs to actors. Put simply, it makes it possible to deal with “subject-verb-object” stuctures, and beyond, at any arbitrary level of complexity, rather than word bags or word distributions which leave a large part of the interpretation to the observer. SOCSEMICS introduced the notion of “semantic hypergraphs” which enable sophisticated computations on claims. An open-source platform called “graphbrain” implements these concepts.
• Achieved the conception of an entirely novel text processing approach, which has been validated and surpasses several key approaches of the relevant information extraction state of the art on relevant tasks, combines different types of AI in an intelligible way, and is available as a platform which already triggered interest by third parties, both academic and non-academic (see above semantic hypergraphs and graphbrain)
• Understanding of the socio-semantic configuration of polarized conversational spaces (i.e. induced by mentions in the Twitter case), which must be distinguished from affiliation spaces (induced by retweets and follows) as they feature both cross-cutting and aligned interactions, such that semantic clusters are not social clusters, thus a counter-example to socio-semantic alignment.
• Description of heterogeneous content accessibility on online platforms through social navigation, in particular, such that the navigational landscape defined by non-personalized recommendations is generally likely to confine users in homogeneous clusters of content. This also emphasizes the contribution of and the variability of the shapes of socio-semantic clusters from the user viewpoint.
• Understanding of the socio-semantic configuration of polarized conversational spaces (i.e. induced by mentions in the Twitter case), which must be distinguished from affiliation spaces (induced by retweets and follows) as they feature both cross-cutting and aligned interactions, such that semantic clusters are not social clusters, thus a counter-example to socio-semantic alignment.
• Description of heterogeneous content accessibility on online platforms through social navigation, in particular, such that the navigational landscape defined by non-personalized recommendations is generally likely to confine users in homogeneous clusters of content. This also emphasizes the contribution of and the variability of the shapes of socio-semantic clusters from the user viewpoint.