Research On Text Emotion Classsification Model Based On Deep Learning

Because of its ability to collect current noteworthy events,Weibo is a social network platform with a large number of participants and a large amount of information,through which the public can express their views,so it is widely used and gradually gaining an important position in the broader field of social media.Microblog comments are often high-dimensional,with sparse semantics and usually contain strong emotions.Emotional integration of the data will have an important impact on network security control.Through the emotional analysis of microblog comments on a certain event,people’s emotions towards the event can be clearly understood.In the task of text emotion classification,three commonly used methods are based on traditional emotion dictionary,machine learning and deep learning.However,the traditional dictionary and machine learning methods over-rely on manual methods to establish dictionary and feature dependence.Therefore,this paper chooses the deep learning method,which can pay more attention to the context information of text and deep network training,and its accuracy is significantly improved compared with the previous two methods.However,when solving emotion classification problems,convolutional neural networks,cyclic neural networks and their variants are faced with some challenges.For example,it is difficult for a single model to obtain high accuracy and assign the same weight to words.From their internal structure and advantages and disadvantages,the main research work of this paper is as follows:(1)Emotions are divided into "positive","negative" and "neutral" for common text emotion classification tasks.In order to more accurately analyze the emotion of microblog text,this paper further divides the text emotion into seven emotional categories,including happy,sad,shock,etc.,through more detailed classification,so as to better analyze the emotion.To achieve better applications.(2)In view of the challenges faced by the current text emotion classification model,such as poor contextual information relevance and missing information,this paper applies the improved Transformer model to the task of emotion classification.Firstly,adding Minikeys variables into Transformer’s multi-head attention mechanism can avoid the problem of sparse data and reduce gradient disappearance and gradient explosion.Meanwhile,the IN standardized method is used to make model training more consistent with natural cognition,so as to extract full-text feature information.Then use Text CNN for convolution pooling operation,obtain local features of the text,integrate it with full-text feature information captured by Transformer,which introduces Minikeys variable,and input it into the full connection layer to realize emotion classification,which effectively improves the classification accuracy of microblog text.The accuracy of model classification is improved by 2.79% compared with unimproved Transformer-Text CNN.(3)Since the setting of Batchsize,Epoch and Dropout is very important in model training,which directly affects the performance of the model,this paper sets multiple groups of parameters for comparison experiment to select the optimal parameter values and obtain the optimal model accuracy.Through the experiment,the values of the three parameters are set as 48,20 and 0.5,respectively.The experiment results are the best.