Exploring Hate Speech Detection in Multimodal Publications

By Raul Gomez, Jaume Gibert, Lluís Gómez, Dimosthenis Karatzas. 2019

308 citations recorded by the source corpus.

In this work we target the problem of hate speech detection in multimodal publications formed by a text and an image. We gather and annotate a large scale dataset from Twitter, MMHS150K, and propose different models that jointly analyze textual and visual information for hate speech detection, comparing them with unimodal detection. We provide quantitative and qualitative results and analyze the challenges of the proposed task. We find that, even though images are useful for the hate speech detection task, current multimodal models cannot outperform models analyzing only text. We discuss why and open the field and the dataset for further research.

Subfields: Artificial Intelligence, Information Systems, Multimodal Learning, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

In Multimodal Learning

In Safety, Robustness & Interpretability

Citations in the corpus

A Multimodal Framework for the Detection of Hateful Memes (2020, background): “[9] introduce MMHS150K, a multimodal dataset of tweets consisting of both image and text and which are manually annotated for hate speech.”
The Hateful Memes Challenge: Detecting Hate Speech in Multimodal Memes (2020, background): “[25] introduced a larger (and arguably noisier) dataset for multimodal hate speech detection based on Twitter data, which also contains memes and which would probably be useful as pretraining data for our task.”
Directions in Abusive Language Training Data: Garbage In, Garbage Out (2020, background): “Although some work has addressed the task of detecting hateful images [118, 121], this lead to the creation of a publically available labelled training dataset in only one case [51].”
TweetBLM: A Hate Speech Dataset and Analysis of Black Lives Matter-related Microblogs on Twitter (2020, background): “(Jha and Mamidi, 2017) proposed the datasets with classes belonging to sex-ist, racist, benevolent sexism, and hostile sexism.”
Exploring Visual Engagement Signals for Representation Learning (2021, background): “This area includes (1) detecting cyberbullying and hate speech [29, 71, 20, 40], (2) identifying emotions [43, 1, 59, 62, 81], (3) understanding rhetoric and intentions [36, 37, 70, 30, 31, 88, 74, 45, 34].”
Winoground: Probing Vision and Language Models for Visio-Linguistic Compositionality (2022, background): “…understanding implausible scenes [13], exploitation of language and vision priors [11, 27], single word mismatches [64], hate speech detection [26, 32, 41, 92], memes [39, 75], ablation of one modality to probe the other [22], distracting models with visual similarity between images [7, 33],…”
Introducing CAD: the Contextual Abuse Dataset (2021, background): “This compares favourably with other abusive content datasets (Gomez et al., 2020; Fortuna and Nunes, 2018; Wulczyn et al., 2017), especially given that our taxonomy contains six primary categories.”
Implicitly Abusive Language – What does it actually look like and why are we not getting there? (2021, background): “Indeed the community is aware of this form of abuse and there have been several attempts for multimodal analysis (Singh et al., 2017; Yang et al., 2019; Gomez et al., 2020).”
Directions in abusive language training data, a systematic review: Garbage in, garbage out (2020, background): “Although some work has addressed the task of detecting hateful images [75, 77, 78], this has not led to publicly available datasets.”
Can Machines Help Us Answering Question 16 in Datasheets, and In Turn Reflecting on Inappropriate Content? (2022, result): “This definition aligns with definitions of previous work detecting hate speech [Gomez et al. 2020] and offensive product images [Gandhi et al. 2020].”