Research And Implementation Of Domainoriented Text Sentiment Analysis Technology Based On Deep Learning

Sentiment analysis or opinion mining is a computational study of people's subjective views on the entity and its subsidiary or specific definitions.The start and rapid development of this field is consistent with the development of social media.For example,domestic social media such as We Chat,Weibo,QQ space,Douyin,etc.Research uses social media information for medical research,of which psychiatry is one of its hot areas.This thesis focuses on the field to solve a technical problem and an application problem.Technical problem: Google released a language model BERT based on a deep twoway Transformer.This model is completely based on the attention mechanism.At the same time,the network that can be calculated in parallel and trained can be deeper,but the model is measured on small data sets in some fields.Not better than the accuracy of neural networks based on RNN or CNN.Based on this,this thesis proposes an improved network model method based on emotion symbol weights,which aims to explore the linguistic mechanism and characteristics of ironic expression in the field of emotion analysis,that is,the construction of the emotion symbol library has different emotion symbols in the text.The importance of extracting effective features from ironic text,including vocabulary,syntax,emoji,and emotional characteristics.The corresponding weights are automatically learned through the neural network training data set,and then the Transformer model is combined with the calculated weights to finally output the emotional level.This method improves the stability and accuracy of the Transorfmer model in this study.Application problem: The hazards of perinatal depression have been identified as one of the "three killers of female mental health".Strengthening the diagnosis and screening of perinatal depression is an effective measure to reduce the incidence.It is currently used in perinatal practice in clinical practice.The diagnosis of depression is basically the same as the diagnosis of depression.It is roughly divided into three categories: scale screening,blood biochemical examination,and imaging examination.This traditional method is relatively subjective and has cultural adaptation problems.Supplemented with a lot of manpower and material resources.Based on social media(We Chat circle of friends)combined with perinatal depression screening,this paper describes document-level emotion classification.By applying the Transformer model based on emotion symbol weights,the user's emotion feature extraction,feature combination,and Analysis of the results,and finally the screening results.An attempt was made to exploratively apply a deep learning network model to screen for perinatal depression.Through experiments and traditional scale surveys,it was initially demonstrated that deep learning can be used for perinatal depression screening and reached the EPDS postnatal depression scale.Screening results are similar.This method has a more objective response to the real situation of perinatal users than the current manual screening methods,which reduces the communication cost between doctors and patients to a certain extent and improves the efficiency of early screening of perinatal depression.The research in this thesis has positive significance for sentiment classification tasks in specific fields,and at the same time provides new ideas and methods for documentlevel sentiment analysis.