ERI models and crisis coordination

This deliverable outlines the development of various models under WP1 that can be used to identify and monitor food safety (emerging) risks, along with the initial design of the interactive dashboard that integrates output of the different models.  

Pipeline models with structured data have been developed to predict the occurrence of pesticides residues in maize, poultry and lentil supply chain, as well as mycotoxins occurrence in the peanut supply chain. These pipeline models can serve as a good foundation for further improvement with better quality data and different machine learning algorithms.  

Different models were developed with unstructured data to identify signals linked to emerging food safety risks. After testing different topic modelling algorithms, BERTopic model was selected for future development under HOLiFOOD context. The weak signal miner was developed to identify terms or concepts within a text corpus that are potentially underrepresented but may gain significance in the future. Three EMM filters were developed, each tailored to one of the HOLiFOOD supply chains (i.e. maize, chicken and legume). The future work will be refining text mining models and assessing the performance of various models with experts. Furthermore, knowledge graphs will be developed to extract insights from literature. 

The initial mock-ups of the interactive dashboard were built based on the current outputs of models with structured data and unstructured data. It is currently under review by the model developers to see if the information is correctly represented in the dashboard. The dashboard will be further adjusted and optimized based on the feedback received from the modeler and insights gained from the living labs.