Scholarlink

This study provides an integrated legal–statistical–technological analysis of social-media defamation. We combine (1) doctrinal comparative review (Indonesia ITE vs. selected U.S./EU standards), (2) an empirical appellate corpus of 45 decisions (2018–2024) and 120 expert interviews, and (3) computational experiments on 5,000 labeled social-media posts to benchmark defamation detection models. We apply χ2 tests to measure cross-jurisdictional enforcement variation, estimate a regression model of prosecution likelihood (jurisdiction type, political sensitivity, reputational-harm score), and evaluate ML pipelines (TF–IDF + BERT embeddings; SVM, Random Forest, Transformer fine-tuning) using stratified 5-fold CV. Key results: significant cross-jurisdictional disparities (χ2, p < 0.001), deep-learning pipelines reach F1 ≈ 0.90 (Precision 0.92, Recall 0.88) on held-out data, and a regression shows criminal regimes increase prosecution odds by β≈0.34 (p < 0.01). We propose targeted policy reforms: narrow criminal provisions, adopt restorative-justice pathways for low-severity cases, and require human-in-the-loop AI with notice/appeal for platform takedowns. These findings demonstrate both the promise and limits of automated moderation when balanced with procedural safeguards.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Defamation in the Age of Social Media: Legal Responses to Digital Reputation Crises

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Haider Abdulkareem Alobaidi,
  • Zahraa Ghazi Sadiq,
  • Rafid Ali Laftah Hamad,
  • Hameed Salim Alkabi,
  • Waleed Nassar,
  • Dmytro Kocherev

摘要

This study provides an integrated legal–statistical–technological analysis of social-media defamation. We combine (1) doctrinal comparative review (Indonesia ITE vs. selected U.S./EU standards), (2) an empirical appellate corpus of 45 decisions (2018–2024) and 120 expert interviews, and (3) computational experiments on 5,000 labeled social-media posts to benchmark defamation detection models. We apply χ2 tests to measure cross-jurisdictional enforcement variation, estimate a regression model of prosecution likelihood (jurisdiction type, political sensitivity, reputational-harm score), and evaluate ML pipelines (TF–IDF + BERT embeddings; SVM, Random Forest, Transformer fine-tuning) using stratified 5-fold CV. Key results: significant cross-jurisdictional disparities (χ2, p < 0.001), deep-learning pipelines reach F1 ≈ 0.90 (Precision 0.92, Recall 0.88) on held-out data, and a regression shows criminal regimes increase prosecution odds by β≈0.34 (p < 0.01). We propose targeted policy reforms: narrow criminal provisions, adopt restorative-justice pathways for low-severity cases, and require human-in-the-loop AI with notice/appeal for platform takedowns. These findings demonstrate both the promise and limits of automated moderation when balanced with procedural safeguards.