| At present,the scale of Chinese Internet users is growing.As a large-scale online social media,microblog provides a platform for Internet users to express their emotions and views.In this era of receiving a huge amount of information quickly every day,some unexpected and major events often spread rapidly and get widely discussed.In the highly liberalized network environment,platforms such as microblog,which spread short text messages,are easy to produce public opinion crisis.It is of great practical significance to understand people’s emotional tendency towards hot events,so as to control the public opinion crisis timely and effectively.In the past research on emotional analysis of hot events,the analysis is often not comprehensive enough.Most of them are analyzed by establishing an emotional dictionary,which can not recognize the hidden meaning in the text,and the effect of the algorithm is not good enough.These papers seldom study the theme and emotional changes of hot events,and lack of analysis of specific events.Taking the network public opinion of recent hot events as an example,this paper crawls the relevant comment information of four representative hot events from June 2020 to June 2021 on the microblog,and regards it as the emotion classification data set of microblog hot events.In addition,the data set also includes the public data sets of microblog emotion classification: Nlpcc2013 and Nlpcc2014.For other expressions other than Chinese characters in comments,previous papers directly eliminated these contents.This paper believes that these data can not only be regarded as noise,and a Chinese thesaurus of non-Chinese expression transformation with a high proportion of data set is constructed to replace the non-Chinese expression which may contain emotional tendency.Due to the high labor cost of manual annotation,it may lead to the shortage of data volume and diversity.Therefore,this paper proposes to enhance the text data before emotional analysis.In terms of algorithm research,the native model of BERT is improved based on machine learning method and deep learning method.The emotion classification model is constructed by combining the last layer of ALBERT with other classification models,and verified on three data sets.The results show that ALBERT-LR model and ALBERT-SVM model have high accuracy and short training time.Especially for emotion classification through ALBERT-SVM,the accuracy rate,the recall rate and F1 value are increased by one to three percentage points compared with the original ALBERT model.In addition,this paper selects the cotton incident in Xinjiang for specific analysis.And combine TF-IDF and ALBERT to improve the LDA topic model.Then,through the discovery of the topic of the comments,we can get the topic that microblog users pay most attention to the event and find the corresponding attitude to the topic.Finally,the emotional value of each comment is combined with the praise number of comments to calculate the daily emotional mean and monthly emotional mean.This paper also analyzes the evolution process of events,and carries out theme analysis and emotion analysis of hot events in stages.The results show that people hold a positive attitude for a relatively long time.In the active period of the event,people’s emotion fluctuates obviously,but there is no phenomenon that the average daily emotion is too low.It can be seen that most people treat the event rationally.This paper studies the evolution process of hot events and people’s attitudes by constructing theme analysis and emotion analysis models,so as to provide reference for management departments to understand the development of public opinion,and then help them effectively prevent and control public opinion crisis. |
