Research And Implementation Of Public Opinion Early Warning Model Based On Image Content In Network Community

The network community has the characteristics of convenience,rapidity and randomness in data providing and disseminating.Meanwhile it is the rally point of social public opinion.The analysis and early warning of information in the network community is endowed with a great importance in helping relevant managers make timely and accurate judgments,curbing the spread of harmful rumors effectively,protecting public safety and maintaining social stability.As all we can see,image is the largest carrier of data dissemination on the Internet.The spread speed and influence of image is far greater than that of text.But at present,the technology of public opinion early warning which based on network community image is still in its infancy.This paper takes the images in the network community as the research object,focuses on the network community of colleges and universities.Combining the characteristics of emergency events in universities,we design the classification system of public opinion elements in the network community of universities and sum up the seven types of public sentiments such as ideological and political,natural disasters,public health,security,school management,national religion,special students and 38 categories of images which related to them.The first stage is that collecting relevant images from network communities and open source image libraries through web crawler technology and labeling them to create the image database.And then,we build a classification model of images in a network community,a text recognition model of images,and an emotional orientation analysis model of images.Experiments show that the accuracy of the three models is 89.31%,98.43%and 98.53%respectively.Combined with the results of three analysis models,we design a public opinion early warning model which based on image content.The experiment was conducted in the more than 60 thousand topics which related to university internet public opinion,there are 81.67%of the classification factors of internet public opinion elements,75.03%of the public opinion warning precision rate and 95%public opinion warning recall rate.