The Research And Application Of Sentiment Analysis And Opinion Leader Discovery Techniques In Social Media

In recent years,social media platforms in China have experienced tremendous development,attracting many scholars to undertake research on social media-related issues.Among these research directions,sentiment analysis and opinion leader detection are regarded as the most representative fields in social media analytics,and have become one of the main focuses for researchers.Social media not only contains user profile and behavior information,but also involves a large volume of text data.However,current sentiment analysis methods struggle to adapt to the domain specificity,complexity,and noise in textual content from social media,while current opinion leader detection algorithms fail to fully consider both user profile and behavior information.To address these challenges,this paper proposes improvement and optimization measures for sentiment analysis methods and opinion leader detection algorithms in social media,and applies these algorithms to specific events,which will be discussed in detail below.Firstly,to address the issue of users expressing emotions through emojis and emoticons,and the existence of specific domain words and internet slang,this paper proposes a comprehensive sentiment dictionary construction method.To achieve this,we use Word2 Vec technology to improve SO-PMI,and use this technology to expand domain-specific and internet-slang-related emotion words.Finally,we compile the emotion dictionaries for emojis and emoticons,domain-specific words,etc.,to obtain a more comprehensive sentiment dictionary.Secondly,to address the problems of lacking appropriately tagged corpora and low sentiment classification accuracy for popular events,this paper proposes a selfsupervised sentiment classification model.This model effectively combines the advantages of sentiment dictionary and deep learning methods through three-way decisions.The model utilizes the emotion dictionary label to assist the deep learning model in supervised learning.By integrating the optimized word vectors derived from the sentiment dictionary into the attention mechanism,the model becomes more capable of capturing the location and emotional tendency of emotion words in the text,resulting in a better understanding of the emotional semantics and an improved accuracy in emotion recognition.Thirdly,to address the problem that current community opinion leader detection algorithms do not comprehensively consider users’ initial influence and leadership indices,this paper proposes an improved opinion leader detection algorithm.This algorithm comprehensively considers user attributes and behaviors,taking into account their initial influence as well as their leadership index within the community,thereby enabling us to more fully identify community opinion leaders and better understand and manage community sentiment.Fourthly,in response to the issue that the emotions of users who forward and comment on posts can be influenced by the emotions expressed in the original post,a behavior-based sentiment analysis method is proposed.Finally,the sentiment analysis method and opinion leader discovery algorithm proposed in this paper are applied to conduct user sentiment-oriented analysis of the "Argentina wins the World Cup" event.