Today,Weibo becomes a popular social media and information dissemination platform in China.In 2016,with the Rio Olympic Games and Wang Baoqiang’s divorce drama,the important role Weibo played in information transmission was significantly highlighted.September’s statistics in 2016 also showed that,the count of Weibo’s active users reached 297 million and the average count of daily active users reached 132 million,which were 34% more and 32% more compared to the numbers of last year,correspondingly.By publishing,forwarding,commenting and liking/disliking blogs,users express their thoughts on people/events and sometimes exchange ideas with others.By analyzing the blogs that people forward and comment,we can quickly get the hottest concerns of the society and the public opinion towards specific affairs.This can be a helpful reference for decision makers.For enterprises,our focus of this study,it’s also common to conclude clients’ opinions towards their products and services based on the related blogs people publish,forward and comment on.In this work,the Emotion Analysis System based on Weibo blogs is proposed and it’s proved to be able to calculate the current public opinions towards specific companies or products fast and accurately,which has important application value for rapid decision making,crisis public relations and public opinion guidance.This manuscript is mainly about concluding Weibo users’ emotional conditions by analyzing their comments on blogs.The works we’ve done can be divided into the following parts.Firstly,a web crawler is implemented to collect blogs and comments for the specific company over the entire internet.Second,the data collected in the first step is pre-processed by filtering based on stop words and segmentation.Thirdly,three classifiers(based on SVM,convolution neural network and LSTM neural network)are trained using the word vectors generated from the blogs/comments by word2 vec.Then an efficient and practical algorithm is proposed by comparing all candidates’ accuracy,recall rate,F1 value,run time,etc.The fourth part is the design of the UI.The effectiveness of the algorithm was verified using the public database COAE2013 and the results showed that the classifier based on LSTM neural network had the best performance.Furthermore,a classifier using an optimized stacked LSTM on the same dataset did show a 1% better performance over the classifier with a regular LSTM.We also experimented several popular methods of categorizing short text like Weibo blogs.In addition,the web crawler helped us collect a huge amount of corpuses.And for selected Weibo accounts,we also used the crawler to get all the comments under their related posts.Finally,using the LSTM model,the Weibo-based Emotion Analysis System was built being visualizable and easy to use. |