The Research And Application Of Emotional Analysis In Social Media Text

With the continuous advancement of mobile Internet and social media platforms,these platforms have become a widely used means for network users to share information and obtain advice.Social media comment texts come from all walks of life and analyzing the large amount of information contained in these texts is of practical significance for information prediction,public opinion monitoring,and user decision-making.This thesis employs microblogs,Twitter,and other domestic and foreign social media as data support and conducts research at both sentence-level and aspect-level granularity.The main research areas are:(1)For sentence-level sentiment analysis tasks,this thesis proposes a hybrid neural network sentiment analysis model that combines character features and word features.The complex sequence structure of the Chinese language and the absence of clear separators in Chinese sentences for word segmentation may lead to loose standards in the word segmentation model.To address this,the thesis proposes a feature extraction method that combines character-level features and word-level features.The model uses the Bi-GRU structure to extract feature relationships that contain context information in the character sequence.It also employs an attention mechanism to focus on important aspects in the character sequence and uses the CNN network structure to extract local feature relationships between words.Different sizes of convolution kernels are employed to obtain local features at varying distances,and the two features are ultimately fused to obtain global feature information.The model was tested on Weibo and other data sets through multiple experiments to provide a thorough analysis.(2)Different aspect words in a sentence will have different effects on the emotional polarity of the text.In order to obtain more detailed analysis results and fully understand the comment information,for fine-grained aspect-level sentiment analysis,this paper proposes a nested attention mechanism based on and Transformer’s aspect-level sentiment analysis model.Traditional fine-grained sentiment analysis usually models the target aspect words and contextual text separately,which cannot effectively obtain the interactive relationship between the two parts of content.The model in this paper calculates the two-way attention weight between the aspect words and the contextual text,To obtain the interaction features between the two,fully consider the interaction between each word in the context and each word in the aspect word,and get the interaction relationship between the two.At the same time,for the upstream feature extraction task,the powerful feature extraction capability of Transformer-encoder is used,combined with the Bi-LSTM network structure to make up for the lack of sequence relative position information in Transformer,making the model more perfect in feature extraction.The model has been experimented with multiple comparison models on Twitter and other datasets,and achieved good results.