Scholarlink

The exponential growth of tourist-generated content necessitates efficient quality assessment frameworks to address inherent challenges of information reliability and analytical scalability. This study pioneers a Large Language Model (LLM)-driven approach, integrating supervised fine-tuning with low-rank adaptation and structured prompt engineering to enable multi-dimensional quality evaluation. Covering 485,930 reviews from three major platforms—MaFengWo, TripAdvisor, and Ctrip—the framework achieves superior performance (RMSE = 0.56, NDCG@K = 0.88) in generating accurate quality scores and detailed analytical rationales. Spatial-temporal-semantic analyses reveal platform-specific quality patterns: MFW exhibits stable temporal cointegration and prominent spatial centrality, TripAdvisor demonstrates simplified core-periphery structures, while Ctrip presents dynamic multicentricity. Heterogeneous network analysis further identifies the behavioral regularities of high-reliability users through a random-walk algorithm. The study advances tourism informatics by resolving scalability limitations of manual coding while providing actionable insights for platform governance, including targeted moderation and incentive mechanisms. This paradigm highlights LLMs’ transformative potential in operationalizing tourist-generated content quality assessment at scale, bridging theoretical rigor with practical applicability within digital tourism ecosystems.

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

Quality Assessment of Tourist Generated Contents: A Large Language Model Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jialiang Gao,
  • Peng Peng,
  • Christophe Claramunt,
  • Feng Lu

摘要

The exponential growth of tourist-generated content necessitates efficient quality assessment frameworks to address inherent challenges of information reliability and analytical scalability. This study pioneers a Large Language Model (LLM)-driven approach, integrating supervised fine-tuning with low-rank adaptation and structured prompt engineering to enable multi-dimensional quality evaluation. Covering 485,930 reviews from three major platforms—MaFengWo, TripAdvisor, and Ctrip—the framework achieves superior performance (RMSE = 0.56, NDCG@K = 0.88) in generating accurate quality scores and detailed analytical rationales. Spatial-temporal-semantic analyses reveal platform-specific quality patterns: MFW exhibits stable temporal cointegration and prominent spatial centrality, TripAdvisor demonstrates simplified core-periphery structures, while Ctrip presents dynamic multicentricity. Heterogeneous network analysis further identifies the behavioral regularities of high-reliability users through a random-walk algorithm. The study advances tourism informatics by resolving scalability limitations of manual coding while providing actionable insights for platform governance, including targeted moderation and incentive mechanisms. This paradigm highlights LLMs’ transformative potential in operationalizing tourist-generated content quality assessment at scale, bridging theoretical rigor with practical applicability within digital tourism ecosystems.