Applied Research On Multilabel Sentiment Identification And Sentiment Intensity Prediction Of Texts Using Modifier Structure

The psychological disease in the fast-paced modern life is a problem that cannot be ignored.Through Internet channels,we can talk about psychological problems and get professional medical staffs opinions,which are both convenient and user privacy.Therefore,the internet medical service platform has accumulated lots of social psychological texts.Mining emotional semantic information in text,identifying potential latent categories or predicting dimension values in emotional space,can get fine emotional information.It is of great significance to apply emotion analysis technology to social psychological texts.This paper studies the sentiment analysis of the text in the field of psychological disease,and probes into the problem of sentiment label recognition and sentiment intensity prediction in this kind of text.It includes three parts;The first part builds a deep learning model to predict multi-label psychological diseases text sentiment labeling.Firstly,the question transformation method is used to convert the multi-label problem into multiple single-label problems.Second,the dissertation considers three categories of modifiers,negative adverbs,degree adverbs and modal words.Negative words are used to reverse the semantic polarity of sentiment words in the structure,while degree words increase or decrease the positive or negative degree of emotional words,modal words can weaken the sentiment Intensity.Two modification structures are proposed:Modification Structure 1:a single modifier or a combination of modifiers.Modifying Structure 2:a combination of Modification Structure 1 and emotion word.Finally,a multi-scale CNN model is constructed for the modified structure,and the modified structural features are extracted through different scales of convolution kernels to analyze the effect on multi-label prediction.The second part builds deep a learning model to predict the valence and arousal ratings of Text.First,the text of a category of mental disorders of depression is marked with the rating in Valence and Arousal spaces.Then,in the same way,the binding model between multi-scale CNN and LSTM is constructed according to the characteristics of the modified structure,the former extracts features of modified structures through different scales of convolution kernels,and the latter extracts the order relationships between features to study the effect of the modified structure on the prediction of the sentiment intensity of texts.In order to verify the effectiveness of the proposed multi-label sentiment identification and the Valence-Arousal ratings prediction model proposed in this paper,a large number of experiments were carried out in this paper.Using the model proposed in this paper,experiments were carried out on three datasets including the data set of modified structure 1,the data set containing modified structure 2 and the original data set respectively,and compared with the deep learning model without considering the modified structure.Experiments show that compared with the model without considering the modified structure,the proposed model improves the evaluation index of the prediction results in the three data sets;The data set with modified structure 1 is increased more than the original data set;The dataset containing the modified structure 2 is increased more than the dataset containing the modified structure 1.The model proposed in this paper is applied to the construction of "disease inquiry system",and the system uses the model to realize the functions of pushing the description of the user interrogation problem and predicting the emotional value of the user text.The system can determine the sentiment categories and sentiment intensity of the user’s text,which can help the psychiatrist to follow-up medical treatment.