Chinese Food Safety Entity Recognition Based on BERT Model

Food safety has always been a hot topic of concern for humans. Online news, Weibo, and other online media platforms are generating a large amount of text data in real-time. The availability of this data can help us to promptly detect and address food safety issues, reducing the impact of such events while safeguarding human health. In the field of food safety, there is a lack of publicly available named entity datasets, especially in Chinese text. This paper collected news headlines from online media and inspection announcements from government departments to construct a corpus. It also designed a dual-task BERT model to identify food safety risks and extract food risk-related entities, leveraging the mutual relationship between the two to enhance the model's performance. Experimental results show an F1 score of 0.86 for sequence classification and an F1 score of 0.77 for entity recognition, validating the effectiveness and soundness of the model proposed in this paper.