Scholarlink

<p>Recommender systems help users find items, services, or content matching their preferences to enhance their experience. However, they are not effective in the case of insufficient data. In order to get beyond the cold start problem and data sparsity that limit traditional models, an intelligent tourism recommendation system has been developed. The system uses a deep learning method that combines Bidirectional Encoder Representations from Transformers (BERT) semantic embeddings, label encoding for categorical data, and normalization for numerical features. These inputs are integrated into a Deep Factorization Machine (FM) model to capture complex feature interactions. When evaluated on destination, hotel, and restaurant recommendations, results demonstrate that Deep FM outperforms alternatives: Wide Deep and xDeep FM, particularly in cold start and sparse data scenarios.</p>

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

TripRec: tackling cold start and sparsity in tourism recommendation using combined pre-trained bidirectional transformer encoder representations

  • 期刊论文

  • 发表日期：

    2025年12月10日

查看全文
  • Messaouda Fareh,
  • Melyara Mezzi,
  • Houda M’ziane,
  • Chafika Lakrib

摘要

Recommender systems help users find items, services, or content matching their preferences to enhance their experience. However, they are not effective in the case of insufficient data. In order to get beyond the cold start problem and data sparsity that limit traditional models, an intelligent tourism recommendation system has been developed. The system uses a deep learning method that combines Bidirectional Encoder Representations from Transformers (BERT) semantic embeddings, label encoding for categorical data, and normalization for numerical features. These inputs are integrated into a Deep Factorization Machine (FM) model to capture complex feature interactions. When evaluated on destination, hotel, and restaurant recommendations, results demonstrate that Deep FM outperforms alternatives: Wide Deep and xDeep FM, particularly in cold start and sparse data scenarios.