Scholarlink

<p>Food safety surveillance requires data systems that link hazards, products, establishments, geographic locations, and recall histories. This study presents FoodSafeKG, a knowledge graph-enhanced framework for food safety data integration and prospective risk analytics. FoodSafeKG integrates FDA recall records and a food adulteration dataset spanning 2012–2024 into a graph with 31,071 entities and 76,559 relationships across five entity types. For temporal risk prediction, we derived graph-based features from information available before the prediction window, trained models on 2012–2019 records, and evaluated establishment-level recall risk in a held-out 2020–2022 period. Logistic regression achieved an AUC-ROC of 0.8227 and an average precision of 0.4107 after excluding direct recall-count features to reduce target leakage. Descriptive analyses showed geographic concentration in California, Illinois, and Texas, with hazard patterns dominated by product contamination, misbranding or unreported allergens, and Listeria. FoodSafeKG supports integrated querying, pattern discovery, and leakage-aware temporal prediction, demonstrating the value of knowledge graph methods for food safety surveillance and risk analytics.</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

A knowledge graph-enhanced framework for food safety data integration and prospective risk analytics

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月5日

查看全文
  • Qiteng Zhang,
  • Zhijian Xu,
  • Wenkang Li,
  • Xiaohua Li,
  • Bin Fu

摘要

Food safety surveillance requires data systems that link hazards, products, establishments, geographic locations, and recall histories. This study presents FoodSafeKG, a knowledge graph-enhanced framework for food safety data integration and prospective risk analytics. FoodSafeKG integrates FDA recall records and a food adulteration dataset spanning 2012–2024 into a graph with 31,071 entities and 76,559 relationships across five entity types. For temporal risk prediction, we derived graph-based features from information available before the prediction window, trained models on 2012–2019 records, and evaluated establishment-level recall risk in a held-out 2020–2022 period. Logistic regression achieved an AUC-ROC of 0.8227 and an average precision of 0.4107 after excluding direct recall-count features to reduce target leakage. Descriptive analyses showed geographic concentration in California, Illinois, and Texas, with hazard patterns dominated by product contamination, misbranding or unreported allergens, and Listeria. FoodSafeKG supports integrated querying, pattern discovery, and leakage-aware temporal prediction, demonstrating the value of knowledge graph methods for food safety surveillance and risk analytics.