Analyzing Connections Between User Attributes, Images, and Text

This work explores the relationship between a person’s demographic/psychological traits (e.g., gender and personality) and self-identity images and captions. We use a dataset of images and captions provided by N ≈ 1350 individuals, and we automatically extract features from both the images and captions. We identify several visual and textual properties that show reliable relationships with individual differences between participants. The automated techniques presented here allow us to draw interesting conclusions from our data that would be difficult to identify manually, and these techniques are extensible to other large datasets. Additionally, we consider the task of predicting gender and personality using both single modality features and multimodal features. We show that a multimodal predictive approach outperforms purely visual methods and purely textual methods. We believe that our work on the relationship between user characteristics and user data has relevance in online settings, where users upload billions of images each day.