Challenge: Existing systems for detecting hateful memes lack sensitivity to subtle differences in memes that are vital for correct hatefulness classification.
Approach: They propose to construct a hatefulness-aware embedding space through retrieval-guided contrastive training to identify hatefulness based on data unseen in training.
Outcome: The proposed system outperforms existing models on the HatefulMemes dataset with an AUROC of 87.0 and improves contextual understanding across domains.

Similar Papers

A Context-Aware Contrastive Learning Framework for Hateful Meme Detection and Segmentation (2025.findings-naacl)

Copied to clipboard

Challenge: Empirical experiments show HateSieve surpasses existing LMMs in performance with fewer trainable parameters .
Approach: They propose a framework to enhance detection and segmentation of hateful elements in memes by creating a triplet dataset and an Image-Text Alignment module.
Outcome: HateSieve features a new framework that creates semantically correlated memes and generates contextual embeddings for accurate meme segmentation.
Caption Enriched Samples for Improving Hateful Memes Detection (2021.emnlp-main)

Copied to clipboard

Challenge: Existing methods for classifying memes are difficult to perform, with human accuracy only about 85% . recent state-of-the-art models perform considerably less accurately, achieving up to 64.73% accuracy.
Approach: They propose to use an off-the-shelf caption generator to capture the first image and overlayed text.
Outcome: The proposed tool improves classification accuracy for unimodal and multimodal models . the proposed tool can be used to model the contrast between image content and overlayed text .
Robust Adaptation of Large Multimodal Models for Retrieval Augmented Hateful Meme Detection (2025.emnlp-main)

Copied to clipboard

Challenge: Large Multimodal Models (LMMs) have shown promise in hateful meme detection, but they face limitations like sub-optimal performance and limited out-of-domain generalization capabilities.
Approach: They propose a robust adaptation framework for hateful meme detection that enhances in-domain accuracy and cross-domain generalization while preserving the general vision-language capabilities of LMMs.
Outcome: The proposed framework outperforms larger agentic systems in detecting hateful memes under adversarial attacks while maintaining the general vision-language capabilities of LMMs.
Deciphering Hate: Identifying Hateful Memes and Their Targets (2024.acl-long)

Copied to clipboard

Challenge: a growing body of research has focused on the negative aspects of memes in high-resource languages like Bengali . a new dataset for Bengali hateful memes is designed to detect their targeted entities .
Approach: They propose a multimodal dataset that analyzes the modality of memes and compares them with other datasets.
Outcome: The proposed dataset outperforms state-of-the-art datasets on Bengali hateful memes . the proposed dataset is generalizable on other low-resource hateful memes datasets compared with baselines based on the proposed model .
Uncertainty-Guided Modal Rebalance for Hateful Memes Detection (2024.acl-long)

Copied to clipboard

Challenge: Existing methods for integrating hate information from different modalities ignore the modality uncertainty caused by the contribution degree of each modality to hate sentiment.
Approach: They propose an Uncertainty-guided Modal Rebalance framework for hateful memes detection . they propose to combine cross-modal fusion features with unimodal features .
Outcome: The proposed framework produces state-of-the-art performance on four widely-used datasets.
MUTE: A Multimodal Dataset for Detecting Hateful Memes (2022.aacl-srw)

Copied to clipboard

Challenge: social media has enabled information propagation at unprecedented rate, but also generated malign content, such as hateful memes . a multimodal hate speech dataset is used to study the impact of hateful content on society . current studies focus on monolingual memes, but existing models cannot provide accurate inferences based on code-mixed captions a study on Bengali memes shows that joint evaluation of visual and textual features significantly improves the hateful data classification .
Approach: They propose to use a multimodal hate speech dataset to detect hateful memes . they use monolingual captions in English and Bengali to analyze the content .
Outcome: The proposed dataset shows that evaluation of visual and textual features significantly improves the hateful memes classification compared to unimodal evaluation.
Text or Image? What is More Important in Cross-Domain Generalization Capabilities of Hate Meme Detection Models? (2024.findings-eacl)

Copied to clipboard

Challenge: Existing studies show that only the textual component of hateful memes enables the multimodal classifier to generalize across domains while the image component proves highly sensitive to a specific training dataset.
Approach: They propose to use only the textual component of hateful memes to generalize across different domains while the image component is highly sensitive to a specific training dataset.
Outcome: The proposed model performs similarly to hate-meme classifiers in a zero-shot setting, while the introduction of meme’s image captions worsens performance by an average F1 of 0.02.
DISARM: Detecting the Victims Targeted by Harmful Memes (2022.findings-naacl)

Copied to clipboard

Challenge: DISARM is a framework that uses named-entity recognition and person identification to detect all entities a meme is referring to and then incorporates a novel contextualized deep neural network to classify whether the meme intends to harm these entities.
Approach: They propose a framework that uses named-entity recognition and person identification to detect all entities a meme is referring to and incorporates a novel contextualized deep neural network to classify whether the meme intends to harm them.
Outcome: The proposed framework outperforms 10 unimodal and multimodal systems and reduces error rate of harmful target identification by 9 % absolute over baseline systems.
MemeIntel: Explainable Detection of Propagandistic and Hateful Memes (2025.emnlp-main)

Copied to clipboard

Challenge: Existing methods for label detection and explanation generation have been limited in understanding complex issues . identifying propaganda and hate in memes is essential for combating misinformation and minimizing harm .
Approach: They propose an explanation-enhanced dataset for propaganda memes in Arabic and hateful memes on English to solve these tasks.
Outcome: The proposed model outperforms the current state-of-the-art in label detection and explanation generation.
Deciphering Implicit Hate: Evaluating Automated Detection Algorithms for Multimodal Hate (2021.findings-acl)

Copied to clipboard

Challenge: Imlicit hate content has unusual syntax, polysemic words, and fewer markers of prejudice, e.g., slurs . multimodal content is harder to detect than unimodal content, such as memes .
Approach: They evaluate the role of semantic and multimodal context for detecting implicit and explicit hate . they find that all models perform better on content with full annotator agreement .
Outcome: The proposed model outperforms other models on implicit and explicit hate detection tasks because of its lower propensity towards false positives.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations