Heuristic and LLM-Based RoI Extraction for Turkish News Pages Türkçe Haber Sayfalarinda Keşifsel ve LLM Tabanli ?Ilgi Bölgesi Çikarimi


Toka D., Sahin U., Gok A., Kumas O., TORAMAN Ç.

34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/siu71813.2026.11636889
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Anahtar Kelimeler: content extraction, HTML processing, information retrieval, large language models, news extraction, Turkish news pages
  • Orta Doğu Teknik Üniversitesi Adresli: Evet

Özet

This study addresses the extraction of news content regions (Region of Interest, RoI) from Turkish news pages using both rule-based and large language model (LLM)-based methods. A heuristic container element is first selected from the HTML DOM; then either this container or the full page body is fed to the LLM under different input types. A 2 × 3 factorial design crossing two scope levels (container/body) with three input types (HTML / Flow-Filtered / Flow-AllText) is evaluated on 100 news samples using weighted token-multiset F1 and ROUGE-L F1. Results show that HTML-based inputs consistently outperform flow-based alternatives across all models, with body + HTML achieving the highest quality and container+HTML offering the best quality-cost trade-off.