Conference Agenda

This study examines the evolution of socio-economic relationships in response to the algorithmic transformations of digital platforms, with a particular focus on TikTok and ChatGPT. The objective is to analyze the extent to which the algorithmic structures of these platforms influence user interactions, access to information, and consumption behaviors. Building upon established theoretical frameworks, it has been observed that the transition from traditional social networks to the algorithmic selves characteristic of TikTok (Bhandari & Bimo, 2022; Boccia Artieri & Donato, 2024) has evolved into a more individualized and mechanized dynamic in platforms such as ChatGPT (Gutiérrez, 2023), ultimately materializing the concept of self-datafication (Couldry, 2020). Within this framework, the study critically investigates how these socio-political transformations have redefined economic structures, not only shaping new consumption practices but also constraining the agency and relational dynamics of traditional economic actors.

The research is structured into two methodological phases. The first phase entails a platform ethnography of TikTok and ChatGPT, aimed at examining their technical and structural configurations as well as the social and cultural practices they foster. The second phase consists of a more extensive empirical investigation, conducted through the administration of semi-structured questionnaires to a sample of 110 Italian participants, to assess the impact of algorithmic mediation on consumer behavior.

The anticipated findings seek to illustrate that the technological transformations of digital platforms are not merely advancements in computational design but manifestations of broader political and cultural shifts with profound implications for economic dynamics, decision-making processes, and spending patterns.

This research proposes a hybrid analytical framework for examining how design and content attributes shape discussion network structures. The framework integrates quantitative and qualitative content analysis, social network analysis, and path modeling, providing a structured approach to uncovering how these attributes interact and influence discussion dynamics.

The framework employs path analysis using Hayes’ (2022) PROCESS macros, enabling a detailed examination of mediation effects. Within this analysis, we assign attributes as mediators, while network measures serve as the dependent variables. This approach allows us to assess the existence of different causal paths and, together, illustrate an underlying mechanism.

The framework is demonstrated through a case study of discussion networks derived from online news comments. These networks function as deliberative spaces and are structured as reply networks, where a directed and weighted tie is formed whenever a commenter responds to another comment, linking the responding commenter-node to the top commenter-node. Each news story generates a distinct discussion network.

In this study, network structures are calculated for each story and incorporated into a model alongside its measured attributes. The model also includes control variables such as news story length, topic, and publishing year to account for potential confounding factors. This approach allows us to trace the path between online environment design, content features, and discussion structure The analysis results provide new findings that address a long-standing knowledge gap regarding audience engagement with data journalism, propose underlying causal mechanisms, and contribute empirical insights into deliberation theory.

Power is everywhere: social reality is based on power relationship whether formal or informal. A city does not have a power per se but concentrates people with power relations. Despite numerous studies of city power, there is a research gap on multilevel city power, and on the link between the power of individuals that transforms into power of cities in global economic networks.

We can wonder how micro-level processes such as human interactions shape the power of cities on global scale? This concerns a question of trust and power in social networks (Castells, 2013), in particular, the networks of those who take key decisions in cities, but also those who take decisions concerning places of the world outside the city (like headquarters). Are these two kinds of networks linked to each other like it is suggested in “Local buzz, global pipelines” (Bathelt et al., 2004)? These networks of decision makers create power asymmetries in cities, control urban governance structures and globalize cities from top-down and bottom-up, thus reinforcing the influence of a city on a world scale.

We conduct a literature review based on the topic modeling method, to underline the main associations between different scales and concepts. Using Google Scholar, we construct several corpuses of literature. Analyzing the networks of terms and highlighting words’ associations allows us to determine the common points and differences between diverse power discourses, and to identify the key elements one can use to define a multilevel/multiscale city power.

References:

Bathelt, H., Malmberg, A., & Maskell, P. (2004). Clusters and knowledge: local buzz, global pipelines and the process of knowledge creation. Progress in human geography, 28(1), 31-56.

Castells, M. (2013). Communication power. Oxford University Press, USA.

During periods of threat, such as with the COVID-19 pandemic, the public looks to officials to understand the situation, its causes, and its ramifications. Among the communication strategies employed by public entities is the use of causal narratives, a highly condensed, structured method of imparting information that describes events in terms of the conjunction of a cause and an effect. As narratives with particular structures have the potential to relay essential, clarifying, and possibly life-saving information to a lay audience, understanding narrative structure is important for developing an evidence base to inform future crisis response. However, an individual’s conception of a situation is rarely informed by a single speakers’ narratives but rather by a welter of messages from different agencies that may be imperfectly remembered and lumped together into an aggregate narrative. Thus, there is a need to examine the broader discourse constructed by articulated causal narratives by responding agencies to the public.

In this study, we reconstruct the semantic network of causal narratives utilized by organizations in public communications during the COVID-19 pandemic, characterizing the underlying structure of the discourse, and assessing the degree to which officials en masse provided clear, consistent, and actionable information on the evolving crisis. A key feature of our work is our evaluation of large language models (LLMs) for extracting complex semantic network features from massive corpuses of heterogeneous text. In addition to comparing the efficacy of LLMs to human performance, we also experiment with both novel and known methods for improving LLM accuracy.

We present a methodology to analyze semantic (co-occurrence) networks of academic papers in a given subfield with complementary and different entropy-based measures to obtain a rigorous and quantitative assessment of unique language usage at concept-level and text-level.

After cleaning each text through tokenization, stop-word removal, and lemmatization, and by following the semantic network creation procedure, we create an adjacency matrix representing co-occurrences between words appearing in the same sentence, conducted by AutoMap (Carley et al., 2013) and ORA (Carley et al., 2014). Each word will then have an associated probability distribution over its neighbors, allowing us to compute its Shannon entropy H(w) and reinterpret it as a quantitative reserve for entropy-based calculations.

We focus on three entropy-based measures. After the implementation of respective normalization steps at each measure, first, each node w has a word-level entropy H(w) reflecting the unpredictability of its immediate neighbors. Second, a local excess entropy Elocal(w) subtracts H(w) from log2(∣N(w)∣), where ∣N(w)∣ is the number of neighbors for node w, thus comparing the observed entropy to a uniform distribution among those neighbors. Third, a global excess entropy Eglobal(w) subtracts H(w) from log2(∣V∣), taking the entire vocabulary into account for its maximum entropy baseline. Finally, by averaging these word-based excess entropies across all nodes, we obtain a single text-level aggregate measure of structural predictability for each text. These complementary entropy-guided quantities provide concept-based and text-based systematic comparisons of linguistic and conceptual structuredness, uniqueness, and comparative basis for texts and concepts that characterize a given academic subfield.