On the Complementarity of Images and Text for the Expression of Emotions in Social Media

TL;DR

This paper investigates whether images and text in social media posts provide complementary emotional information, introducing a new annotated corpus and models to analyze their combined and individual contributions to emotion understanding.

Contribution

It presents a novel annotated corpus of multimodal Reddit posts and evaluates the necessity of both modalities for emotion and stimulus detection tasks.

Findings

Text alone predicts most image-text relations.

Multimodal models excel in anger and sadness detection.

Image-only models are best for objects, animals, and food.

Abstract

Authors of posts in social media communicate their emotions and what causes them with text and images. While there is work on emotion and stimulus detection for each modality separately, it is yet unknown if the modalities contain complementary emotion information in social media. We aim at filling this research gap and contribute a novel, annotated corpus of English multimodal Reddit posts. On this resource, we develop models to automatically detect the relation between image and text, an emotion stimulus category and the emotion class. We evaluate if these tasks require both modalities and find for the image-text relations, that text alone is sufficient for most categories (complementary, illustrative, opposing): the information in the text allows to predict if an image is required for emotion understanding. The emotions of anger and sadness are best predicted with a multimodal model, while text alone is sufficient for disgust, joy, and surprise. Stimuli depicted by objects, animals, food, or a person are best predicted by image-only models, while multimodal models are most effective on art, events, memes, places, or screenshots.