Research On Short Text Emotion Classification Based On Deep Neural Network

The era of big data and artificial intelligence has promoted the revolution of Internet social media,and at the same time prepared the conditions for the high-quality development of online text.Text sentiment analysis is one of the core tasks of natural language processing.It is dedicated to extracting emotional semantic features by analyzing,learning,and inducing the context information of unstructured text,so as to mine the sentiment tendency expressed in the subjective text.Therefore,how to accurately obtain the hidden value information of the text content from a large number of complicated information is the central link of scientific research in the field of text sentiment analysis.However,traditional text sentiment analysis mainly faces two challenges: one is that the selection of text features is easily affected by the subjectivity of the consciousness of the text feature extractors,and it is impossible to dig deep into effective hidden information;the other is due to the key extracted based on traditional methods Words cannot effectively establish the attribute relationship between key keywords,and the extracted features are not representative enough,and the form is too simple.In recent years,deep learning has achieved good results in many fields including natural language processing.Based on this,in the field of sentence-level text sentiment analysis based on deep learning,this paper integrates the advantages and characteristics of the current mainstream neural network model to construct a fusion model based on deep learning,and combines an attention mechanism(Attention Mechanism),Principal Component Analysis(PCA),Naive Bayes Classifier(NBC)and other machine learning methods,two models are proposed:1.Aiming at the problem that the traditional convolutional neural network extracts the emotional features of the text is relatively simple,and it cannot balance the contradictory relationship between the current feature pooling method and the feature vector dimension is too high and the semantic information is retained.A CNNpbc model(Convolutional Neural Network plus Bayes Classifier)based on traditional convolutional neural network and PSGD(Partial Sampling Gradient Descent)model update algorithm is proposed.CNNpbc uses parallel double pooling operation of k-max + avg pooling in the pooling layer to better preserve the semantic features of the text.The model uses the PSGD algorithm to ensure the stability of the training process and improve the convergence speed of the model and improve the accuracy of classification.(The model was published in IEEE-ICICT2019 in 2019)2.Aiming at the current problem that the deep neural network model is relatively independent and cannot fully aggregate the advantages of each model while not fully considering the keyword influence factors.The keyword analysis in specific target sentiment analysis is integrated into sentence-level text sentiment classification,and a MATT-CNN + BiGRU fusion model is proposed.The model uses the three major elements of word vector,part of speech,and position in the attention mechanism to construct the embedding matrix of emotional word vectors.It can not only take advantage of the multi-attention CNN's strong ability to extract n-gram features and extract local features of target keywords,but also combine the BiGRU model with a relatively simple structure and can take into account the global features of the text to fully consider the advantages of the contextual semantic information of the words.,Breaking the shackles of the long-term development limitations of a single model,and providing a new idea for the development of sentence-level text sentiment analysis.(The model was published in the EI journal "Future Internet" in 2019)...