| With the prosperity of WEB 3.0,social media has been developing fast.By the end of November 2016,Facebook has more than 1.4 billion registered users,Twitter's monthly active users have reached 310 million,and the number of Sina Weibo's monthly active users also has reached 260 million.A large number of active users are spontaneously contributing to the extraction,creation,dissemination of news and information every day,so massive current events spread quickly on social media sites.Therefore,it is significant to detect events from the social streams such as Sina Weibo.Event detection and tracking is an event-based analysis and utilization of information.At present,there is not much research on this filed.Moreover,due to the tweets' characteristics such as fragment and noisy,existing event detecting ways are not performing well.In this paper,we propose an effective method by combing shallow learning feature and deep learning feature to detect event.LDA is adopted as the shallow learning feature,and Word2Vec as our deep learning feature.By combing the two features,the deep context information and topic information will be fully utilized.Then,based on events we detected,we propose an event hotness prediction method and a sentiment orientation judgment method.In this paper,we construct a corpus,which is grabbed from Sina Weibo and covers a time span of more than two months.By combining LDA-based shallow learning feature and Word2Vec-based deep learning feature,and using Random Forest method to predict each tweet's event label,We can use some parameters such as tweets' forwarding,comments,likes,is Big V or not and so on,to construct the linear regression model,calculating and predicting the event heat.We also put forward a sentiment orientation analysis method based on the evaluation of the object's expectation.In order to verify the effectiveness of our methods,we evaluate the methods on our human annotated corpus.Experiment results show that our methods perform very well. |
PDF Full Text Request