Sentiment Analysis Of Public Safety Events In Microblog Comments Incorporating Features At Multiple Levels

With the rapid development of the Internet and the popularization of web2.0technology,social media applications have become popular all over the world at an alarming rate.The ever-expanding social media service provides a carrier and platform for the spread of public opinion.Nowadays,microblog has become an important way for domestic people to obtain information,transmit and express emotional opinions in real time.However,its open,convenient and spread-fissionable characteristics have also buried hidden dangers for the rapid outbreak and fermentation of public opinion.In particular,destructive events related to the safety of the public often trigger a tipping point at the beginning of public opinion and quickly focus the attention of public.Because of the asymmetry of information and the emotional inflammatory nature of speech,these unhealthy microblog will cause dread social panic and disturb public order without management.Sentiment analysis and public opinion control of microblog have already attracted the attention of scholars in the early days.Various studies aim to promote the intelligent supervision of online public opinion through the automatic recognition of sentiment opinions in microblog comments.However,research on sentiment analysis of public security is scarce.We mainly propose a relatively comprehensive feature system at multiple levels to improve accuracy of microblog sentiment analysis for public safety events.First of all,there are different potential topics in the life cycle of microblog.And the microblog comments under the topic have different topic attributes.Simultaneously,emotions expressed by users are synergistic when they are in the same period,the same topic,or the topic attribute.We extracts the evolutionary features according to the characteristics of public opinion in the field of public safety,and extracts the features of the topic attributes of microblog comments by analogy with fine-grained sentiment analysis in the field of product reviews.Then,we draws on previous studies to extract features of microblog content from punctuation,part-of-speech,syntax,and sentiment words in comments.Finally,because people in different categories have similar emotion expression rules,we mines the user’s potential attributes under the premise of combining the user’s basic attributes,so as to extract the user’s portrait features.After constructing a feature system based on the 20 extracted features above,we use XGBoost algorithm,random forest algorithm and support vector machine algorithm to conduct sentiment classification experiments,which can identify non-negative emotions and negative emotions in microblog comments.After constructing a feature system based on the 20 extracted features,the XGBoost algorithm is used to construct a classifier to conduct sentiment classification experiments,which can identify non-negative emotions and negative emotions in microblog comments.At the same time,the random forest algorithm and the support vector machine algorithm are used as the comparisons.The experimental results show that the 20 features all have a positive effect on the accuracy of microblog sentiment analysis for public safety.The features of microblog comments such as punctuation,syntax,and sentiment words have the highest contribution to model recognition,followed by features of public security opinion and user portrait such as life cycle,topic,time periods,personality tendency and following number.The sentiment analysis model of XGBoost,random forest and support vector machine has an accuracy of more than 80% and an operating speed of less than 20 s for the sentiment analysis of microblog reviews in the field of public safety.Among them,the XGBoost model performs best with an accuracy of 85.42% and the running speed is 16.93 s,which can efficiently and accurately identify non-negative and negative comments in the sentiment analysis task of microblog comments in the field of public security.