Social Media Image Popularity Prediction Based On Deep Learning

With the rapid development of Internet technology,many social media platforms such as Facebook,Instagram,and We Chat,etc.have sprung up,leading to massive multimedia images.Users visit multimedia images,favorite and forward the information of interests,which boosts up the spread and popularity of social media images.Therefore,the research on the prediction of social media image popularity has become a current hot issue,which is of great significance in the fields of content recommendation,online marketing,and media advertising.This thesis focuses on the prediction of social media image popularity based on deep learning.The main contents are as follows:To fulfill the immediacy requirements of image popularity prediction,this thesis first proposes a method for predicting the popularity of social media images based on MSDNet.This method uses MSDNet to extract image information features and uses multi-scale convolution to obtain high-resolution feature maps.After that,it uses the Text CNN model to extract text information features and employs convolution kernels with different sizes to extract key information in the sentence to better obtain the local correlation of the text.Finally,the XGBoost algorithm is utilized as a regressor to complete the image popularity prediction.Experimental results show that the proposed method outperforms the existing ones in terms of prediction accuracy and calculation efficiency.In addition,to improve the accuracy of image popularity prediction,this thesis proposes a Res Net-based popularity prediction method for social media images.This method uses Res Net to extract image information features,which can effectively handle the phenomenon of gradient disappearance during deep network training and ensure the expression ability of output features.Following this,the New-Stacking algorithm is utilized as the regressor to effectively avoid the over-fitting problem caused by the limited amount of data and to improve the accuracy of image popularity prediction.Experimental results show that this method significantly improves the accuracy of image popularity prediction compared with the existing image popularity prediction methods.In summary,this thesis targets the requirements of both immediacy and prediction accuracy of social media image popularity prediction and proposes two methods for predicting the popularity of social media images.Experimental results show that the proposed methods effectively achieve the fast and high-precision prediction of image popularity.