
Large Language Models (LLMs) show strong text generation capabilities, yet their emotional consistency in social media contexts remains underexplored. This study examines emotional and semantic behaviors of four models: Gemma, Llama 3, Llama 3.3, and Claude in continuation and response tasks using climate change discussions from Twitter and Reddit. Results show that while all models maintain high semantic coherence, they systematically attenuate negative emotions, often shifting them toward neutral or even positive tones. Compared to human-authored content, LLM outputs exhibit reduced emotional intensity and a consistent preference for neutral, rational expression. Despite differences between continuation and response tasks, semantic similarity remains high across models. These findings highlight a general tendency of LLMs toward emotional moderation, offering insights for their deployment in social media and human-AI interaction design.
Wikipedia is one of the most visited websites globally, yet its role beyond its own platform remains largely unexplored. In this paper, we present the first large-scale analysis of how Wikipedia is referenced across theWeb. Using a dataset from Common Crawl, we identify over 90 million Wikipedia links spanning 1.68% of Web domains and examine their distribution, context, and function. Our analysis of English Wikipedia reveals four key findings: (1) The topics of Wikipedia articles referenced on the Web differ from those cited on Reddit and those prominent within Wikipedia's own link structure, (2) Wikipedia is most frequently cited by news and science websites for informational purposes, while commercial websites reference it less often. (3) The majority of Wikipedia links appear within the main content rather than in boilerplate or user-generated sections, highlighting their role in structured knowledge presentation. (4) Most links (95%) serve as explanatory references rather than as evidence or attribution, reinforcing Wikipedia's function as a background knowledge provider. While this study focuses on EnglishWikipedia, our publicly released WEB2WIKI dataset includes links from multiple language editions, supporting future research on Wikipedia's global influence on the Web.
Social media use is linked to low fertility desires. However, we know little about the discourses surrounding family formation that people consume online. We analyze 219,127 comments on 668 short videos related to fertility and parenthood from Douyin and Tiktok in China, South Korea, and Japan, a region famous for its extremely low fertility level, to examine the topics and sentiment expressed. BERTopic model is used to assist thematic analysis, and a large language model QWen is applied to label sentiment. We find that comments focus on childrearing costs in all countries, utility of children, particularly in Japan and South Korea, and individualism, primarily in China. Comments from Douyin exhibit the strongest anti-natalist sentiments, while the Japanese and Korean comments are more neutral. Short video stances and content creators' identity significantly influence the responses, alongside regional socioeconomic indicators, including GDP, urbanization, and population sex ratio. This work provides one of the first comprehensive analyses of online discourses on family formation via popular algorithm-fed video-sharing platforms in regions experiencing low fertility rates, making a valuable contribution to our understanding of the spread of a low-fertility discourse online.
The Arab Spring was a cascade of protests and uprisings across the Middle East and North Africa that toppled governments and escalated into prolonged conflicts in others. Collective memories of these events evolve unevenly across political, cultural, and linguistic contexts. Drawing on archived content from the Arabic and English Wikipedias (2011-2024), this study provides a generalizable framework for studying long-term collective memory processes on Wikipedia. We propose multilingual measures of salience, deliberation, contextualization, and consolidation and evaluate them using English and Arabic Wikipedia articles about the Arab Spring. Article size and wikilinks indicate sustained salience in both languages, though along different trajectories. Temporal clustering of wikilink inclusion reveals Stable, Debated, and Forgotten content, marking zones of consensus, contestation, and exclusion. Interlanguage links show limited cross-lingual alignment, with only a minority shared between editions. English pages consistently link back to the Arab Spring summary, while Arabic pages often do not, yielding uneven integration of the phenomenon into the encyclopedic network. These findings demonstrate Wikipedia's role in long-term memory work and highlight cross-lingual biases relevant to multilingual knowledge systems.
This study examines the relationship between moral framing and fundraising outcomes, encompassing both monetary and social support, through an analysis of 14,088 GoFundMe campaigns. Focusing on three moral foundations (care, fairness, and ingroup loyalty), we quantified the presence of these frames within campaign appeals. Our results show that the association between moral framing and campaign success varies significantly across fundraising categories. Specifically, negatively framed appeals emphasizing the harm experienced by the help-seeker were associated with a higher volume of donations and comments in the Emergency and Memorial categories, yet related to fewer donations in animal-related campaigns. Highlighting harm was not associated with a higher average donation amount per donor. Ingroup loyalty framing was positively associated with both donation volume and supporter engagement. This research extends the literature on moral communication in crowdfunding and offers practical implications for the design of online fundraising platforms to better facilitate prosocial interactions between fundraisers and supporters.