Researchers have adapted a large language model (LLM) framework, LLM-SC, to better extract information from reviews of sacred heritage sites, specifically focusing on the Pashupatinath Temple in Nepal. This adaptation involved creating a specialized 12-type entity taxonomy, including eight new types for sacred heritage aspects, and refining extraction prompts for GPT-4o. The system processed 2,500 Tripadvisor reviews, yielding over 11,000 entity, aspect-sentiment, and relation records, with the new sacred-heritage types comprising the majority of extracted entities. Human evaluation indicated provisional scores of 41.0% precision and 21.8% recall for entities, and 42.6% for end-to-end aspect-plus-polarity. AI
IMPACT This research demonstrates a method for improving LLM performance on niche, culturally specific data, potentially enabling better analysis of heritage sites and similar domains.
RANK_REASON The item is an academic paper detailing a domain-specific adaptation of an LLM framework and its evaluation. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
