algopol

Understanding information sharing and filtering on the Internet

Algopol aimed at proposing a typology of information sharing behaviors and at understanding the role of algorithms in the ranking of Internet content. Project members adopted both an empirical and a comparative approach by contrasting different online spaces and, as much as possible, different countries (diversely focusing on French-, German-, English-speaking communities).

More precisely, Algopol distinguished three main online contexts where users produce and diffuse content:

1. public self-expression spaces such as blogs, micro-blogs (first and foremost Twitter)
2. platforms where users address a limited, selected audience, such as social networking sites (featuring Facebook) 
3. collaborative knowledge creation platforms, such as wikis (including Wikipedia).

The project endeavored to describe information sharing and authority phenomena across all spaces, asking similar questions while getting differentiated answers.

Our consortium proposed this project in 2011 after identifying two blind spots in the literature on informational and interactional dynamics on the Internet :

• First, the state of the art on web network morphogenesis (structure and dynamics) had been steadily growing over the previous years – it was also the main topic of a previous ANR grant called Webfluence. Yet, few studies had been specifically connecting topology and content, notably in the context of authority phenomena and the formation of implicit groups of users sharing or consuming similar information.
• Second, a noticeably increasing body of research was devoted to content diffusion, however addressing this issue mostly through a quantitative lens, without qualitatively deconstructing individual information sharing behaviors or differentiating across use cases and contexts. The role of algorithms was moreover significantly under-estimated.

Here, the comparative and qual-quant approach of Algopol has been key in proposing an integrated answer to these two main classes of questions.

Computational social study of online content sharing and authority phenomena

Methodologically, Algopol relied on three pillars :

• (i) the extraction and construction of databases and the development of empirical protocols enabling the in vivo study of online information sharing,
• (ii) the sociological analysis of interaction, information hierarchization and recommendation processes, both in a qualitative and quantitative framework,
• (iii) the understanding and development of algorithms related to the reconstruction of these dynamics, both in terms of typologies as well as through numerical and formal models.

More broadly, this project shed light on the crucial combination of social science and computer science approaches. On the one hand, we could apply a series of data mining techniques on large data sets (« Big Data »), articulated around graph theory (network topology, dynamic and heterogeneous graphs, cluster detection) and natural language processing (nominal group detection, semantic categorization). On the other hand, beyond traditional qualitative investigation methods (interviews, corpus analysis), we could develop original and innovative protocols making it possible to assemble so far unavailable data connecting large user-centric datasets with socio-demographic and other qualitative characteristics, thanks to the development of a viral recruitment app on Facebook.

Main results

Algopol demonstrated the importance of understanding online information spaces as a scaffolded architecture of realms obeying distinctive ranking and visibility principles. We could for instance show that information sharing practices are strongly differentiated depending on whether users interact within public expression spaces such as blogs and micro-blogs, semi-public platforms such as social networking sites, or collaborative knowledge production communities such as wikis. In all cases, informational landscapes are singularly influenced by the shape of interactions, and singularly fragmented with respect to user properties – be they semantic (topics of interest, socio-demographic features, etc.) or relational (position in the network, within communities).

The issue of the diversity, plurality or neutrality of algorithms appeared as a key question to understand the transformation and even economic challenges of online ecosystems. Here, Algopol will provide a robust preliminary to the forthcoming Algodiv research project (2016-19) devoted to the description of socio-informational diversity on the web, as a result of the joint behavior of users and algorithms.

Organizational details

Algopol was a fondamental research project coordinated at CAMS (UMR CNRS/EHESS) by Camille Roth and involving Orange Labs (Dominique Cardon), LIAFA (UMR CNRS/Paris VII) and Linkfluence (Guilhem Fouetillou), also receiving support from Centre Marc Bloch Berlin (UMIFRE CNRS/MAE/BMBF/HU). The project officially started in July 2012 and lasted for about three years. It benefitted from an ANR grant amounting ot €428k, for a total cost of around €1.1m.