Exploring Visual Engagement Signals for Representation Learning
This work addresses the challenge of bridging low-level visual information with high-level social interactions for representation learning, offering a novel method for social media analysis.
The paper tackled the problem of using visual engagement signals from social media as supervisory signals for representation learning, and demonstrated that their VisE approach effectively improves performance on subjective downstream tasks like emotion recognition and political bias detection.
Visual engagement in social media platforms comprises interactions with photo posts including comments, shares, and likes. In this paper, we leverage such visual engagement clues as supervisory signals for representation learning. However, learning from engagement signals is non-trivial as it is not clear how to bridge the gap between low-level visual information and high-level social interactions. We present VisE, a weakly supervised learning approach, which maps social images to pseudo labels derived by clustered engagement signals. We then study how models trained in this way benefit subjective downstream computer vision tasks such as emotion recognition or political bias detection. Through extensive studies, we empirically demonstrate the effectiveness of VisE across a diverse set of classification tasks beyond the scope of conventional recognition.