Baptiste Kotras: Measuring online speech. Traces, mechanisms and regimes of public opinion on the web

“Know what people think”. This slogan, displayed on the homepage of the British company Brandwatch’s website, sums up the promise of this new industry dedicated to analysing public opinion online: finally knowing what people think, without even having to ask them. Since the mid-2000s, a range of agencies, start-ups and software publishers have indeed been offering to gauge public opinion based on the millions of daily posts on social media: tweets, blog posts, forum messages and even comments in the online press. This vast pool of data is thus utilised by a wide variety of players and tools to measure everything from the reputation of major brands and companies to reactions to an advertising campaign or a televised debate, or even emerging consumer trends; what they all have in common is their focus on the unsolicited opinions of internet users. This approach has, moreover, enjoyed considerable success for several years: in 2015, ten years after the first companies in this field were established, the social media analysis software industry generated $2.2 billion in revenue, according to the consultancy Markets and Markets, which also forecasts that this market will grow to $17.9 billion by 2019.

Distancing itself both from the positivism that surrounds these technologies and from a constructivist critique that remains external to its subject matter, this thesis seeks to describe the emergence of a new type of mediation that harnesses conversational traces from the social web to measure public opinion, thereby renewing the epistemological and political content of this category. Starting from the hypothesis that there is a plurality of forms of opinion and of the mechanisms that measure it, we thus examine an extremely varied set of models, developed by start-ups and innovative companies. We are thus initially interested in attempts to redefine the notion of representativeness through the sampling of mobilised and influential audiences. Secondly, we examine the model that appears to be gaining the upper hand in this market, based on software aimed at the exhaustive and continuous indexing of traces of online conversation. By measuring opinion as a continuous flow of singular events, these tools break definitively with the idea of representativeness and revitalise a framework for understanding opinion as a collective and discursive activity, attentive to the media and relational dynamics of online discussion. This indexical regime of opinion knowledge thus moves away from quantifying majorities and minorities towards measuring mobilisations and shifts in public opinion.

Members of the jury

• Loïc Blondiaux, Professor at the University of Paris I Sorbonne
• Dominique Cardon, Associate Professor at the University of Paris-Est Marne-la-Vallée
• Patrice Flichy, Professor Emeritus at the University of Paris-Est Marne-la-Vallée (Director)
• Alexandre Mallard, Research Director at Mines ParisTech
• Cécile Méadel, Professor at the University of Paris II Panthéon-Assas
• Sylvain Parasie, Senior Lecturer at the University of Paris-Est Marne-la-Vallée (co-director)