Research On Classification Of Agricultural Products' Public Opinion Information Based On Deep Learning

As a big agricultural country,Agricultural products are necessity for us to survive.In recent years,public opinion on the amplification and speculation of agricultural product quality and safety issues has greatly increased the difficulty of agricultural product quality and safety supervision.How to quickly grasp the bias of public opinion and timely distinguish between positive and negative factors in public opinion is extremely important for agricultural supervision and management departments.Therefore,the classification of public opinion information is the primary task when we analyze public opinion information of agricultural products.This paper proposes a classification method for agricultural product public opinion information,which is improved from three aspects: word vector,convolutional neural network and recurrent neural network,feature fusion and multi-model fusion.An improved study based on word vectors.In the process of constructing word vector in word2 vec,there is no supervision and no document category information.In this paper,the word weight value of TF-IDF is used as the weight of word2 vec word vector matrix.The improved word vector has text distinguishing ability.In view of the premise that word2 vec prescribes words accurately,this paper proposes to generate feature vectors from letter granularity and word granularity simultaneously,this way can broaden the dimension of feature vectors and improve the accuracy of text classification.Improved research based on convolutional neural networks and recurrent neural networks.For the TextCNN convolutional neural network has only a single convolutional layer and its max-pooling layer will lose the lexical order.This paper increases the depth of the network by recurrent convolution and semi-pooling.The improved TextCNN can obtain the relevance among long-distance text.For TextCNN,only one-dimensional convolution is performed on the sentence dimension.It lacks the convolution feature on the word embedding dimension.In this paper,the twodimensional convolution is used to convolve the word vector matrix.The improved TextCNN can obtain the local features of the word vector dimension and broaden the feature dimension.In view of the fact that LSTM cannot extract local features in parallel,this paper combines CNN and BiLSTM,which can obtain local features and obtain global information of text sequences.The experimental results show that all of the improved models improve the performance of the classification.Research on classification of public opinion information based on feature fusion and multi-model fusion.Single model extracts the characteristics of public opinion information insufficiently.This paper enhances the model's ability to extract public opinion information by integrating the shallow features of the model,which can obtain higher-level features with more ability of public opinion information classification.For the insignificant effect of single model on the classification of public opinion information,this paper combines the improved model with the classical model through various strategies.Finally,this paper evaluates the classification effect of the above fusion models.We select the fusion model with the best classification effect appling to the national agricultural product traceability system to classify the public opinion information of “Xiangxi Citrus Slow Sales” in the system.The experimental results show that the fusion model selected in this paper can classify agricultural product public opinion information with high precision,and it has certain reference significance in the field of agricultural product public opinion information classification.