Image Popularity Prediction (IPP) Over Time Using Machine Learning Techniques

Abstract

The destiny of social media images depends upon their popularity: some of the uploaded images/videos get a lot of fame among people while others just get completely unnoticed. Why is this so? This work addresses this question, discusses all the features related to social content that are responsible for its popularity or negligence and also propose a system to predict the popularity of the content for the span of 30 days before actually uploading the content on any social media platform. There are some common features in the social content that gets fame, in this research work we have evaluated the effect of different features on the popularity score of the content. The proposed model predicts the popularity score in the form of number of views for the next 30 days after uploading the content. The content popularity score can be used by companies to improve their marketing strategies, targeting the right audience sagaciously, managing the resources efficiently and making the strategical decisions. In research work, the detailed methodology is discussed to design a model that can perform the task of Image Popularity Prediction (IPP) efficiently. A critical analysis is also performed on the results obtained from single features, combinational features and features obtained by applying different techniques. The best results are achieved by using Linear Discriminant Analysis (LDA) technique, which converts the higher dimensional features into six dimensions. This provides 8.03 value for tRMSE and spearman correlation of 0.76. This research work manifest that the features related to the image context i.e. user features and photo features etc. outperforms other features related to the content of the image.