The recognition of emotion has become an important application of artificial intelligence, which allows systems to understand human affective states from social media content. The fast increase of user-generated multimedia data, in particular images and visual posts, has raised an increasing demand for analysing not only the static emotions but also their temporal evolution, which we call emotion drift. Existing methods mainly focus on emotion classification at a single instance level, while temporal dependencies in long-term user data are often ignored.

In this review paper, a detailed review of deep learning techniques for emotion recognition is presented with a special emphasis on hybrid CNN-LSTM architectures. Convolutional Neural Networks(CNN) are good at extracting spatial features from images, while Long Short-Term Memory (LSTM) networks are good at capturing temporal patterns and sequential dependencies. The integration of these models offers a deeper understanding of emotional transitions in social media environments.