Recommendations in Academic Social Media: the shaping of scholarly communication through algorithmic mediation

Scholarly communication is increasingly being mediated by Academic Social Media (ASM) platforms, which combine the functions of a scientific repository with social media features such as personal profiles, followers and comments. In ASM, algorithmic mediation is responsible for filtering the content and distributing it in personalised individual feeds and recommendations according to inferred relevance to users. However, if communication among researchers is intertwined with these platforms, in what ways may the recommendation algorithms in ASM shape scholarly communication? Scientific literature has been investigating how content is mediated in data-driven environments ranging from social media platforms to specific apps, whereas algorithmic mediation in scientific environments remains neglected. This thesis starts from the premise that ASM platforms are sociocultural artefacts embedded in a mutually shaping relationship with research practices and economic, political and social arrangements. Therefore, implications of algorithmic mediation can be studied through the artefact itself, peoples’ practices and the social/political/economical arrangements that affect and are affected by such interactions. Most studies on ASM focus on one of these elements at a time, either examining design elements or the users’ behaviour on and perceptions about such platforms. In this thesis, a multifaceted approach is taken to analyse the artefact as well as the practices and arrangements traversed by algorithmic mediation. Chapter 1 reviews the literature about ASM platforms, and explains the history of algorithmic recommendations, starting from the first Information Retrieval systems to current Recommender Systems, highlighting the use of different data sources and techniques. The chapter also presents the mediation framework and how it applies to ASM platforms, before outlining the thesis. The rest of the thesis is divided in two parts. Part I focuses on how recommender systems in ASM shape what users can see and how users interact with the platform. Part II investigates how, in turn, researchers make sense of their online interactions within ASM. The end of Chapter 1 shows the methodological choices for each following chapter.

Part I presents a case study of one of the most popular ASM platforms in which a walkthrough method was conducted in four steps (interface analysis, web code inspection, patent analysis and company inquiry using the General Data Protection Regulation (GDPR)). In Chapter 2 it is shown that almost all the content in ASM platforms are algorithmically mediated through mechanisms of profiling, information selection and commodification. It is also discussed how the company avoids explaining the workings of recommender systems and the mutually shaping characteristic of ASM platforms. Chapter 3 explores the distortions and biases that ASM platforms can uphold. Results show how profiling, datafication and prioritization have the potential to foster homogeneity bias, discrimination, the Matthew effect of cumulative advantage in science and other distortions.

Part II consists of two empirical studies involving participants from different countries in interviews (n=11) and a research game (n=13). Chapter 4 presents the interviews combined with the show and tell technique. The results show the participant’s perceptions on ASM affordances, that revolve around six main themes: (1) getting access to relevant content; (2) reaching out to other scholars; (3) algorithmic impact on exposure to content; (4) to see and to be seen; (5) blurred boundaries of potential ethical or legal infringements, and (6) the more I give, the more I get. We argue that algorithmic mediation not only constructs a narration of the self, but also a narration of the relevant other in ASM platforms, configuring an image of the relevant other that is both participatory and productive. Chapter 5 presents the design process of a research game and the results of the empirical sessions, where participants were observed while playing the game. There are two outcomes for the study. First, the human values researchers relate to algorithmic features in ASM, the most prominent being stimulation, universalism and self-direction. Second, the role of the researcher’s approach (collaborative, competitive or ambivalent) in academic tasks, showing the consequential choices people make regarding algorithmic features and the motivations behind those choices. The results led to four archetypal profiles: (1) the collaborative reader; (2) the competitive writer; (3) the collaborative disseminator; and (4) the ambivalent evaluator.

The final chapter summarises the ways in which ASM platforms forges people’s perceptions and the strategies people employ to use the systems in benefit of their careers, answering each research question. Chapter 6 discusses the implications of algorithmic mediation for scholarly communication and science in general. The dissertation ends with reflections on human agency in data-driven environments, the role of algorithmic inferences in science and the challenge of reconciling individual user’s needs with broader goals of the scientific community. By doing so, the contribution of this thesis is twofold, (1) providing in-depth knowledge about the ASM artefact, and (2) unfolding different aspects of the human perspective in dealing with algorithmic mediation in ASM. Both perspectives are discussed in light of social arrangements that are mutually shaped by artefact and practices.