Forestry Text Classification Based on BERT and KNN

With the rapid development of mobile Internet, the number of various texts on the network is exploding. It is a common demand to classify them. However, in actual production and life, it is often necessary to classify texts in a specific field, such as forestry texts. Use BERT to train the training text, get the text vector of each document, and then average to get the center value of the text vector of each type of document. Based on the idea of KNN, we can judge which kind of text's central vector is closer to each document in the test corpus, so as to achieve the goal of classification. Through experiments, the classification effect of the method based on BERT and KNN idea is better than that of the method using naive Bayes. Using naive Bayes for classification, although the classification effect is not too bad, in order to pursue the accuracy of classification, deep learning models such as the combination of BERT and KNN model are better.