Food contamination and food poisoning are presenting a substantial safety risk to consumers worldwide. In the era of information quantity and availability, the potential of social-media data has attracted increasing attention from relevant government regulatory agencies, food companies, and consumers. This paper takes text data from online media as a research object and innovatively proposes a new type of food text-mining technology based on the associated attention mechanism to quickly screen for potential food safety issues. First, we used the mutual information between each review Chinese word segment(CWS) and label to calculate the correlation score between each word and food safety hazards. Then, the attention score in supervised deep learning was combined in order to assess whether foods sold online may be unsafe for consumers. We compared the method in this paper with existing text-mining methods on food-safety-related datasets and found that the proposed method performs markedly better than the benchmark model, achieving an accuracy rate of 96.95\(\%\). A team of food safety experts also performed a food risk assessment on the prediction results produced by the proposed model, and experimental results showed that the proposed tool can markedly reduce the time required to screen for food safety risks. This study provides a fast and cost-effective food-safety screening method and helps reduce consumer dietary safety hazards.