Content Suggestions Through Tracking of Human Emotions

Abstract

The amount of content that is available to consumers has expanded tremendously with the introduction of social media platforms and online content consumption. The vast volume of material, however, makes it difficult to find and suggest customized content that matches users' emotional inclinations. Additionally, a lot of research has been done on human emotion recognition. However, no research on content recommendation utilizing emotion is mentioned. This study offers a novel approach for content suggesting to overcome this problem. It demonstrates the use of CNN (Convolutional Neural Network) to identify facial expressions and recommend contents. Three models—a custom model developed using CNN and two transfer learning models, VGG16 and ResNet50—are employed and explained in this article. The proposed approaches are assessed using a customized dataset that was produced by combining FER2013, CK+48, and some images collected from online. The custom model performed well with higher accuracy, which is 98.05% with less epochs, as compared to VGG16 and ResNet50, whose frequent accuracy values are 97.22% and 82.36%. Last but not least, the model evaluation has been explored with content suggestions, which is the major goal of this work.