In order to solve the thorny problems in the process of classification of agricultural diseases and insect pests questions, such as fewer public data sets, shorter texts and sparse features, and difficult to learn implicit semantic information, using the hot agricultural investment network as the data source, a data set for the classification of agricultural pests and diseases was constructed, and a deep learning model BERT_Stacked LSTM for the classification of agricultural pests and diseases was proposed. Firstly, the BERT obtained the character-level semantic information of each question, and generated a hidden vector containing sentence-level feature information. Then, stacked long short-term memory network (Stacked LSTM) structure was used to learn the hidden complex semantic information. Experimental results showed the effectiveness of the proposed model. Compared with other comparative models, the model proposed had more advantages in classifying agricultural diseases and insect pests questions. The F1 score reached 95.76%, and it was widely used in public. Tested on the domain data set, the F1 score reached 98.44%, indicating that the generalization of the model was also very good.