Large language models (LLM) in computational social science: prospects, current state, and challenges

Social Network Analysis and Mining, Год журнала: 2025, Номер 15(1)

Опубликована: Март 9, 2025

Язык: Английский

---

The Duality of Persuasion: Between Personalization and Detection

Опубликована: Май 8, 2025

Язык: Английский

---

Overview of the CLEF-2024 CheckThat! Lab: Check-Worthiness, Subjectivity, Persuasion, Roles, Authorities, and Adversarial Robustness

Lecture notes in computer science, Год журнала: 2024, Номер unknown, С. 28 - 52

Опубликована: Янв. 1, 2024

Язык: Английский

---

Detecting Propaganda Techniques in Code-Switched Social Media Text

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, Год журнала: 2023, Номер unknown, С. 16794 - 16812

Опубликована: Янв. 1, 2023

Propaganda is a form of communication intended to influence the opinions and mindset public promote particular agenda. With rise social media, propaganda has spread rapidly, leading need for automatic detection systems. Most work on focused high-resource languages, such as English, little effort been made detect low-resource languages. Yet, it common find mix multiple languages in media communication, phenomenon known code-switching. Code-switching combines different within same text, which poses challenge Considering this premise, we propose novel task detecting techniques code-switched text. To support task, create corpus 1,030 texts code-switching between English Roman Urdu, annotated with 20 at fragment-level. We perform number experiments contrasting experimental setups, that important model multilinguality directly rather than using translation well use right fine-tuning strategy. plan publicly release our code dataset.

Язык: Английский

---

Misleading But Not Fake: Measuring the Difference Between Manipulativeness Discernment and Veracity Discernment Using Psychometrically Validated Tests

Опубликована: Дек. 11, 2023

Misinformation continues to pose a substantial societal problem, but the measurement of misinformation susceptibility has often been done using non-validated tests. Furthermore, research shows that misleading content (implied misinformation) is much more common than outright false (explicit misinformation). However, there very little on predictors belief in implied misinformation, and it unknown if direct are psychologically similar. To address these questions, we ran three preregistered studies (N 1 = 487, N 2 547, 3 490) which developed validated 24-item 12-item Manipulative Online Content Recognition Inventory (MOCRI), test measures person’s ability distinguish between neutral content. This substantially outperforms other known terms its predictive value for people’s correctly identify many kinds We also show different from one another, although they related. Finally, people who score high MOCRI better low performers at discerning manipulative non-manipulative statements (i.e., have “discernment”), this does not necessarily translate discernment quality their sharing decisions, or willingness reply vs. messages. Instead, resilient being manipulated lesslikely share respond both

Язык: Английский

---

Framing Social Movements on Social Media

Journal of Quantitative Description Digital Media, Год журнала: 2024, Номер 4

Опубликована: Май 29, 2024

Social media enables activists to directly communicate with the public and provides a space for movement leaders, participants, bystanders, opponents collectively construct contest narratives. Focusing on Twitter messages from social movements surrounding three issues in 2018-2019 (guns, immigration, LGBTQ rights), we create codebook, annotated dataset, computational models detect diagnostic (problem identification attribution), prognostic (proposed solutions tactics), motivational (calls action) framing strategies. We conduct an in-depth unsupervised linguistic analysis of each strategy, uncover cross-movement similarities associations between features such as pronouns deontic modal verbs. Finally, compare strategies across other social, cultural, interactional contexts. For example, show that is more common replies than original broadcast posts, organizations focus much journalists ordinary citizens.

Язык: Английский

---

REPRESENTATION OF TIKTOK IN U.S. NEWS AFTER THE NO TIKTOK ON GOVERNMENT DEVICES ACT

VNU Journal of Foreign Studies, Год журнала: 2024, Номер 40(4), С. 1 - 21

Опубликована: Авг. 31, 2024

TikTok, the most downloaded application in world for three consecutive years since 2020, has been target of various bans from governments and organisations. Despite widespread media coverage such events, representation TikTok therein remains underexplored by linguists. This study therefore attempts to examine how is linguistically represented U.S. television news after President Joe Biden signed No on Government Devices Act into law. The Systemic Functional Linguistics, elaborated Halliday Matthiesen (2014), was employed as analytical framework. findings revealed that predominantly uses Material processes, followed Relational processes. Mental Verbal processes rank third fourth, while Existential Behavioural have rather low frequency. Regarding their subtypes, Identifying Cognitive are more prevalent negative representations, Attributive Emotive occur often positive representations. These transitivity patterns contribute five main portrayals including “TikTok a scrutiny ban target”, cybersecurity threat”, Trojan horse Chinese Communist Party”, must-have many”, sales driving force”.

Язык: Английский

---

Decoding persuasion: a survey on ML and NLP methods for the study of online persuasion

Frontiers in Communication, Год журнала: 2024, Номер 9

Опубликована: Окт. 2, 2024

The proliferation of digital communication has profoundly transformed the landscape persuasive discourse. Online platforms have amplified reach and impact techniques. However, they also enabled rapid spread manipulative content, targeted propaganda, divisive rhetoric. Consequently, a wide range computational approaches emerged to address multifaceted nature persuasion, detect mitigate its harmful practices. In light this, paper surveys methods for detecting means in communication, focusing on how integrate humanistic knowledge operationalize this construct. Additionally, special emphasis is placed models' explainability, pivotal aspect considering these models are used by institutions influence societal interactions. For analysis, two primary perspectives persuasion defined: linguistic argumentative. approach analyzes specific textual features, allowing highly accountable algorithms based explicit rules. argumentative focuses broader mechanisms, offering greater scalability but often resulting less explainable due their complexity. This tension between model sophistication explainability presents key challenge developing effective transparent detection systems. results highlight spectrum methodologies studying ranging from analyzing stylistic elements explicitly propagandist messages. Our findings challenges using tackle issues misuse: opacity deep learning absence theoretically grounded distinction vicious virtuous persuasion. To challenges, we propose integrating social sciences humanities theories enhance effectiveness ethical robustness interdisciplinary enables more nuanced characterization text, facilitating differentiation through analysis rhetorical, argumentative, emotional aspects. We emphasize potential hybrid that combine rule-based with techniques, as offer promising avenue implementing framework. concludes outlining future including importance multimodal multilingual considerations handling user-generated data growing distinguishing human AI-generated content.

Язык: Английский

---

Holistic Inter-Annotator Agreement and Corpus Coherence Estimation in a Large-scale Multilingual Annotation Campaign

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, Год журнала: 2023, Номер unknown, С. 71 - 86

Опубликована: Янв. 1, 2023

In this paper we report on the complexity of persuasion technique annotation in context a large multilingual campaign involving 6 languages and approximately 40 annotators. We highlight techniques that appear to be difficult for humans annotate elaborate our findings causes phenomenon. introduce Holistic IAA, new word embedding-based annotator agreement metric various experiments using its correlation with traditional Inter Annotator Agreement (IAA) metrics. However, given somewhat limited loose interaction between annotators, i.e., only few annotators same document subsets, try devise way assess coherence entire dataset strive find good proxy IAA tasked different documents languages, which classical metrics can not applied.

Язык: Английский

---

Multimodal Propaganda Detection in Memes with Tolerance-Based Soft Computing Method

Lecture notes in computer science, Год журнала: 2024, Номер unknown, С. 343 - 351

Опубликована: Янв. 1, 2024

Язык: Английский

---