酒精饮料食品安全合规风险信息的自动检测方法

Automated detection of regulatory changes with machine learning to mitigate compliance risks in the alcoholic beverage industry

随着食品安全法规的不断发展,紧跟时事,及时了解条款修订情况对企业确保遵守并降低潜在的风险至关重要。利用计算机技术有助于自动识别监管变更,精简监察流程并能够及时地响应。该研究旨在探讨机器学习算法在食品安全风险管理中的应用。研究提出了一个提高合规风险管理效率和有效性的框架,基于Transformer的双向编码器表示(bidirectional encoder representations from transformers,BERT)——一个预训练的自然语言处理模型,以典型的监督学习模型为基线,被用来识别与特定食品类别潜在食品安全风险相关的新闻信息变化情况。新闻报道中的组织、食物、风险、法规等关键实体被BERT模型自动提取。以酒精饮料为例,结合领域专家提供的标注数据,研究得到了一个微调的(有提高的)BERT模型,该模型可以自动检测与酒精饮料和与之相关的关键实体相关的潜在监管变化。结果表明,相关性预测的F1分值为0.88,实体识别的F1分值为0.60。所提出的方法有可能显著减少手工工作,提高检测监管变化的准确性,最终强化食品企业的合规策略。

With the ever-evolving landscape of food safety regulations,staying abreast of modifications is crucial for enterprises to ensure adherence and mitigate potential risks.Leveraging computational technologies facilitates the automatic identification of regulatory changes,streamlining the monitoring process and enabling timely responses.This study explored the implementation of advanced machine learning algorithms in the context of the food industry,presenting a framework that enhanced the efficiency and effectiveness of compliance risk management.In this work,bidirectional encoder representations from transformers(BERT),a pretrained natural language processing model,was employed to identify regulatory news relevant to potential food safety risks of specific food categories,with typical supervised learning models as baselines.Key entities such as organization,food,contaminant,and regulation reported in the news reports were also automatically extracted by the BERT model.Using alcoholic beverage as an example along with the labeled data provided by domain experts,we obtained a fine-tuned BERT model that can automatically detect potential regulatory change related to alcoholic beverage and the critical entities associated with it.The results showed that the F1-score of relevance prediction was 0.81,and the F1-score of entity detection was 0.60.The proposed approach holds the potential to significantly reduce manual efforts,enhance accuracy in detecting regulatory alterations,and ultimately fortify the compliance strategies of food enterprises.