Social Media Image Classification and Effect of Local Features on Classification Performance
Vedpriya Dongre, Pragya Shukla · Advances in intelligent systems research/Advances in Intelligent Systems Research · 2025
User-generated contents drive social media platforms.Users share heterogeneous data types, i.e., text, image, and video.Among them, a significant amount of data has been posted in image format.Therefore, analyzing and classifying social media-based images have many essential applications.In this paper, we are addressing the issue of social media-based image classification.Social media images have different characteristics than the normal images used in other applications.To simulate this issue, in this paper two scenarios of the experiment have been demonstrated.First, the experiment utilizes three variants of deep learning models to traditionally classify image data.Additionally, the second experiment includes image classification based on textual and visual features.This model is developed using a deep learning concept namely multi-model feature fusion.The implemented models are experimented with a publicly available social media MEME image dataset, which is obtained from Kaggle.Based on the results, we found that only image-visual feature-based classification provides 43% accuracy.On the other hand, when we utilize both textual and visual features of images using the multi-model fusion.Then the classification accuracy has been improved up to 2%.Thus, shortly for social media image classification the multi-model feature fusion techniques are recommended to use.