Scholarlink

Spam comments on social media, especially on posts by public persons, disrupt the dissemination of substantive information. This study presents an enhanced approach to spam detection by integrating two frequently neglected components: emojis and post-comment pairings. Emojis, sometimes overlooked in textual analysis, significantly contribute to expressing user intents. This strategy use stacked post-comment pairings to enhance the capture of context and relevance, rather than concentrating exclusively on comments. Utilizing the SpamID-Pair dataset for Indonesian social media, results indicate that the integration of emoji-text characteristics, post-comment pairs, and ensemble voting techniques enhances detection accuracy and F1-scores. The soft voting ensemble and SVM (RBF kernel) attained superior performance, with carefully selected features yielding further enhancements.

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

Spam Detection on Social Media Using Emoji and Post-comment Features with ML Ensemble Methods

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Maka Sri Lakshmi,
  • Pamidi Srinivasulu,
  • Archana Gelli,
  • S. Gopinath,
  • Vadlapati John Bunyan,
  • Chikatla Praveen

摘要

Spam comments on social media, especially on posts by public persons, disrupt the dissemination of substantive information. This study presents an enhanced approach to spam detection by integrating two frequently neglected components: emojis and post-comment pairings. Emojis, sometimes overlooked in textual analysis, significantly contribute to expressing user intents. This strategy use stacked post-comment pairings to enhance the capture of context and relevance, rather than concentrating exclusively on comments. Utilizing the SpamID-Pair dataset for Indonesian social media, results indicate that the integration of emoji-text characteristics, post-comment pairs, and ensemble voting techniques enhances detection accuracy and F1-scores. The soft voting ensemble and SVM (RBF kernel) attained superior performance, with carefully selected features yielding further enhancements.