| Social media has a large number of users and is a social application with convenient,active,real-time,and interactive characteristics.Weibo is the most typical representative of social media in China,and has become a hub for the dissemination of numerous hot events,as well as a place for netizens to express opinions and share sentiments about social phenomena.This article collects relevant information on specific thematic events,understands netizens’ views and sentimental tendencies towards hot events,analyzes the evolution process of events,enables the government to timely grasp the tense and public opinion development,and helps guide public opinion towards a positive direction.The research work of this paper mainly includes the following contents:1)Data collection and preprocessing on Weibo platform.This paper takes Sina Weibo platform as the collection object,uses web crawler technology to obtain data according to keywords,specific users,specific Weibo ID search and other strategies,collects blog posts and comment data within a certain period of time,and collects the original data for filtering,cleaning,removing stop words,segmentation,tagging and other operations,laying the foundation for subsequent tasks.2)Topic mining to discover hot events.This article focuses on the processed blog data and conducts topic clustering to extract relevant hot events.Combining the generalized Polya Jar model with the topic model,a probabilistic topic model based on semantic reinforcement is proposed.Build a self coding network to train blog data,obtain blog semantic embedded expression information,use the generalized Polya forum model to strengthen vocabulary vocabulary and short text topic semantics,put them into the topic model,cluster out relevant keywords,determine the optimal topic,improve semantic coherence,and better find multiple representative hot events.3)Sentimental analysis,analyzing the sentimental tendencies of netizens.This article selects "Li Yundi’s involvement in pornography" as an example to conduct sentimental analysis on some event data.By improving the text feature processing method,a serial hybrid network short text sentiment classification algorithm(BERT-CNN-Bi LSTM-ATT)is proposed.The pre trained model is used to obtain vocabulary dynamic vectors,and the hybrid network extracts local and global features of short text to analyze the sentimental distribution of netizen views.Experiments have shown that this algorithm solves the problem of sparse short text on Weibo,efficiently and comprehensively processes features,and improves the effectiveness of sentiment classification.4)Evolutionary system,intuitively describing the evolution trend of events and sentiments.Select the "Li Yundi Hot Example" and implement the system based on the above research methods,from the perspectives of events and sentiments.The system uses lifecycle theory to divide the event lifecycle,combines theme discovery and sentiment analysis methods,calculates different sentimental values according to rules,analyzes the context and sentimental evolution process of hot events,and visualizes and depicts the sentimental evolution process of the event in detail. |
