| With the rapid development of Internet,online news has become an important information carrier.More and more people understand current events through browsing news reports.Under this trend,many online news websites have provided emotion voting functions for users,aiming to collect the feelings and opinions of users involved in news reports after they viewed the news.Using the user's emotion voting results to predict the social public's emotional distribution of social events holds great significance.This thesis observes and analyzes the emotion voting results of online news website users,namely readers of news reports,and concludes that readers' emotion voting results have the following two characteristics: at first,readers' emotions are often related to a social hotspot event rather than related to a specific word in the news report;the second is that the reader's emotions are not completely independent,and there is a delicate relationship between each emotion.Based on the conclusion,this thesis jointly models latent topics of news reports and readers' emotions with weighted emotional topic model.At the same time,this thesis uses multi-label learning algorithms to learn the relationship between reader's emotional labels to complete social emotional prediction.The main work of this thesis includes:(1)Social emotion detection based on weighted emotion topic model: Based on the characteristics of reader emotions and social events,this thesis proposes a weighted emotion topic model based on the emotion topic model.The latent topics of the document and the reader's emotion labels are jointly modeled.At the same time,the weight of the training sample documents is evaluated by calculating the emotional entropy of the document,and the relationship between underlying topics of the document and the reader's emotion labels is explored;(2)Social emotion detection based on multi-label learning: This thesis regards the social emotional prediction task as a multi-label learning problem,and uses multilabel learning algorithms to learn the correlation between emotion labels,aiming to improve the accuracy of social emotion detection;(3)Public opinion analysis system based on social emotion detection: This thesis designed and implemented a public opinion analysis system based on socialemotion detection.The system automatically crawls the latest news reports on the Internet,and analyzes social emotions for the crawling news reports.After that,the system displays analysis results to users through web pages intuitively.The system can help users better grasp the trend of events and make timely and correct decisions. |
PDF Full Text Request