Research On Impact Of Information Features Of Media Weibo On Users' Information Behavior Based On ELM

With the continuous upgrading of Weibo platform application scenarios and continuous innovation of communication methods,the media value of Weibo has been greatly highlighted.More and more traditional mainstream media have settled on Weibo,and on the new platform,they use new ways to play the role of vocalizing and recording important social events.At the same time,due to the openness of Weibo information exchange,the same incident will trigger a large number of media reports,and the explosion of homogeneous information is becoming a key issue that hinders media Weibo from competing for user attention.Therefore,how to attract users to participate in information exchange on the characteristics of Weibo information while ensuring their own advantages in traditional mainstream media is becoming an issue worthy of attention and research.In the framework of Elaboration Likelihood Model,this thesis refers to the existing research foundation,and combines the information characteristics of media Weibo.On the central path,it summarizes guidance words,summary sentences,emotional dialogues,interesting symbols,extended forms,text lengths.On the marginal path,it summarizes source authority and source reliability,and transform it into the influencing factors of media microblog user information behavior,and build an ELM-based influencing factor model of media microblog user information behavior.Use Stata and other statistical analysis software to evaluate the model,and perform negative binomial regression empirical analysis to obtain the actual influencing factors.Secondly,based on the actual influencing factors,we continued to refine the feature indicators,using 46 feature indicator categories as input variables,and forwarding and commenting as output variables,and constructing random forest models of forwarding and commenting behaviors,respectively.The "40th Anniversary of Reform and Opening Up" event is used as a training set and the "2019 Two Sessions" event is used as a test set to select and confirm important features and categories that affect the behavior of reposting and commenting.The research conclusions show that in terms of the influencing factors of users' information behavior,among the eight theoretical factors selected in the article,guidance words,summary sentences,interesting symbols,and source authority have significant effects on information behavior,and emotional dialogue,extended form,text length and source reliability have different degrees of influence on retweet,comment and like behavior,while emotional dialogue,extended form,text length,and reliability of the source have a slight effect on forwarding,commenting,and like behaviors,or no effect on a certain type of information behavior.In terms of subdivision feature categories,it was found that the 1-2 subdivision categories in the main influencing factors ranked in the top position in the importance ranking,that is,they have a prominent contribution to the user's forwarding or commenting behavior.Subdivision categories such as guide words and extended forms in categorical variables,subdivision levels such as counting the number of fans and text length in variables,and media microblogs containing this feature category are also more likely to cause forwarding and commenting behavior.