Impact Assessment of Food Safety News Using Stacking Ensemble Learning
Bo Song, Kefan Shang, Junliang He, Wei Qi Yan, Tianjiao Zhang · Advances in transdisciplinary engineering · 2020
Food safety has always been the focus of public concern. Assessment of the impact of food safety news constituents an important job of the government departments. In this paper we present a method using stacking ensemble learning to assess the impact level of global food safety news. The news used for training the assessment model is collected from the Chinese customs. Each of the news articles is annotated with a label ranging from low impact, medium impact and high impact by the customs officials. For base learners in the ensemble learning model, we use Naive Bayesian, Support Vector Machine, XGBoost, FastText, Convolutional Neural Network, LSTM and BERT. A Naive Bayesian-based meta learner is used to integrate the assessment results of the base learners. The proposed method features end to end prediction of news impact using the original news text as input, and it advances the transdisciplinary development of artificial intelligence and risk assessment by improving the accuracy of impact assessment of food safety news compared with traditionally used methods.