Exploring Visual Engagement Signals for Representation Learning

TL;DR

This paper introduces VisE, a weakly supervised learning method that uses social media engagement signals as supervisory cues to improve representation learning for various computer vision tasks.

Contribution

It presents a novel approach to leverage social engagement signals as pseudo labels for training visual models, bridging the gap between visual data and social interactions.

Findings

VisE improves performance on emotion recognition tasks.

Models trained with VisE outperform traditional methods on political bias detection.

The approach demonstrates versatility across multiple classification tasks.

Abstract

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.