Karolina Śliwa
I study how people behave online when the world is on fire.
○
Publications
Using Psycholinguistic Features for Multiplex Community Detection in Temporal Social Networks
Based on three real-world datasets from the #BlackLivesMatter movement, Brexit, and the Christchurch terrorist attack, we derive overlapping temporal snapshots and compare pure interaction networks to two multiplex networks, respectively. In particular, our multiplex networks introduce expressions of grievances and moral framings as additional network layers. For each temporal snapshot, we then apply the multiplex version of Leiden community detection and analyze how the additional network layers influence (i) fragmentation (community count, singleton rate, and size concentration) as well as (ii) temporal persistence across non-overlapping time windows. We found that adding psycholinguistic attribute layers results in more cohesive partitions that can provide deeper insights into social media communication patterns.
Cultural Evolution of Beauty Standards
Who gets to be beautiful, and who decides? The media and fashion industry set aspirational body standards with measurable consequences for body dissatisfaction and eating disorders. Calls for greater diversity have grown louder, and the industry has responded. Using 25 y of data across runway shows, advertisements, and magazine covers, we document a contradiction: Headline measures of diversity have improved, yet the aspirational body is essentially unchanged. Broader casting has emerged, but remains confined to outliers that do not shift the norm. Models cast in these roles are also disproportionately non-White, revealing a symbolic diversification in which the industry satisfies multiple representational demands by concentrating them on already-marginalized individuals rather than changing who counts as aspirational.
Exposure and Adoption of Social Dimensions During the Ukraine War
This paper investigates exposure and adoption of social media users to ten social dimensions during the initial phase of the Ukrainian war. To this end, we analyze a dataset including over 189 million Twitter messages. In order to explore how different social dimensions are adopted and propagated, we derived a communication network from our Twitter dataset and applied a probabilistic approach to model the spread of messages. Our results indicate that messages containing knowledge — in particular those that provide informative and factual content — emerge as key drivers of engagement on Twitter, highlighting the critical role of an informed discourse in shaping the public opinion and social behavior during the conflict.
A Case Study Comparing Twitter Communities Detected by the Louvain and Leiden Algorithms During the 2022 War in Ukraine
This paper presents a case study regarding a comparative examination of the Louvain and Leiden community detection algorithms. The case study was conducted on a real-world communication network consisting of 3,222,623 nodes and 27,423,553 edges. In particular, the network in our case study models the communication between Twitter users during the initial four weeks of the 2022 war in Ukraine. In addition, we also applied dynamic topic modeling in order to examine differences in the detected communities.
An Analysis of Twitter Communities Related to the 2022 War in Ukraine
In this paper, we analyze a dataset including more than 189 million tweets related to the first month of the 2022 war in Ukraine. Our analysis especially focuses on communities of Twitter users and their collective behavior. In particular, we applied the InfoMap community detection algorithm and found on average 44079.63 communities of Twitter users per day. Our behavioral analysis especially focuses on the five largest daily communities. We found that: 1) hashtags played an essential role in framing conversations, 2) communities often publicly called on international organizations such as @NATO or @UN to aid in conflict resolution, 3) anger was the dominant emotion in all communities and 4) negative tweets spread wider than positive ones.
On the Dynamics of Narratives of Crisis during Terror Attacks
We present an analysis of the narratives that emerged on Twitter during four different terror attacks. To this end, we analyze a dataset consisting of more than five million Twitter messages. We use the structural topic model (STM) approach to automatically detect six narratives of crisis. Our findings indicate that i) Twitter users are highly engaged in the dissemination of operational and memorial narratives, ii) emotions and narratives directed towards authority accounts dominate the discourse, iii) the presence of positive authority nodes directly impacts the type of discourse and fuels hopeful memorial narratives to a larger extent than accusations and blaming.