A Food Safety Text Filtering Method Based on Text Classification Techniques

In order to filter out food safety texts from unstructured textual data of various types, we proposed a Chinese food safety text filtering method. Firstly, the data is collected and preprocessed by the web crawler; secondly, the BERT pre-training model is fine-tuned by small-scale data; then the document vector calculation is carried out using a feature extraction method combining TF-IDF values and word vectors of keywords proposed in this paper; finally, the SVM classifier is trained by document vectors to screen out food safety text. The experiments show that the SVM classifier is able to filter out food safety texts from various types of text data with high performance, which basically achieves the expected results.