Research And Implementation Of Image Public Opinion Early Warning Model Based On Graphic And Textual Combination

In the multimedia era with the rapid development of information,images have become the focus of public opinion supervision departments.Network community is an important public opinion communication environment in the multimedia era.The analysis and early warning of network community image public opinion content has important social significance.The mainstream technology of public opinion analysis and early warning of image content is to analyze and early warning sensitive objects by training image classification and recognition model based on the low-level visual characteristics of images.Image features include low-level visual features(color,texture,line,etc.)and high-level semantic features(subject,scene,behavior,etc.).Image early warning technology by classification takes the subject as the recognition object.The image content in the network community environment includes scene,behavior and other event elements,which cannot be expressed in this recognition process,causing poor accuracy and interpretability of image public opinion analysis results in the network community.The fierce discussion of diversified image content among network community users is also an important way to trigger public opinion of image content.Therefore,only by extracting and analyzing the high-level semantic feature content such as image scene and behavior of network community,and integrating and analyzing the diversified image content and heat comments,can we realize the image public opinion early warning technology of network community with higher accuracy and stronger interpretability.This topic designs and implements an image public opinion early warning model based on the combination of graphic and textual mode.Aiming at the problem that the high-level semantic features of images cannot be expressed in the image early warning technology based on recognition and classification.The image content of high-level semantic features of network community images is expressed in the form of text by using imagecaption technology.In view of the diversified image content of explicit text information such as emphasis,guidance and description in the image of network community,this topic designs and implements a multi feature text fusion algorithm based on feature saliency to improve the adaptability of image public opinion early warning technology to the diversified image content of network community.In order to solve the problem of image public opinion caused by the fierce discussion among users around the content of image fusion in the network community,this paper designs and implements the image public opinion early warning algorithm based on graphic and textual combination.Based on textcnn,combined with the prior knowledge of public opinion analysis in network community,analyze the sensitive public opinion elements of text multiple features of images and hot comment text,and improve the accuracy and interpretability of image public opinion early warning technology in network community environment.Finally,this topic designs and implements the network community image public opinion early warning system.The public opinion management department can quickly and effectively observe public opinion through the visual image public opinion demonstration system,guide and intervene the online community public opinion in time,and maintain a harmonious and stable online community environment.In order to verify the performance of image public opinion early warning model based on graphic and textual combination,this topic selects the network community image data set composed of sensitive image data set of network community and Sina Weibo data for comparative experiment.Experiments show that the model integrates and analyzes image content,high-level semantic feature text,diversified explicit text and hot comment text improves accuracy and interpretability of public opinion early warning obviously,better than the mainstream image recognition and classification early warning model.