Findings of the CONSTRAINT 2022 Shared Task on Detecting the Hero, the Villain, and the Victim in Memes

Sharma, Shivam; Suresh, Tharun; Kulkarni, Atharva; Mathur, Himanshi; Nakov, Preslav; Akhtar, Md. Shad; Chakraborty, Tanmoy

doi:10.18653/v1/2022.constraint-1.1

Cited by 6 publications

(12 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Biases. As per the authors, biases found in the dataset are unintentional (Sharma et al, 2022b), and by conducting the study on this dataset we do not intend to cause harm to any group or individual.…”

Section: Discussionmentioning

confidence: 97%

“…A significant influx of memes from online fringe communities, such as Gab, Reddit, and 4chan, to mainstream platforms, such as Twitter and Instagram, resulted in a massive epidemic of intended harm (Zannettou et al, 2018). Several datasets capturing offensiveness (Sharma et al, 2022b). (Suryawanshi et al, 2020), hatefulness (Kiela et al, 2020;Gomez et al, 2020), and harmfulness (Pramanick et al, 2021b), have been curated.…”

Section: Studies On Memesmentioning

confidence: 99%

“…or victims; instead, they serve the purpose of disseminating harmless humour and trivial opinions. Therefore, Sharma et al (2022b) do not presume any implications regarding these connotations and categorise them as 'other', a fourth neutral category, unless expressed otherwise in the meme. A depiction of such a scenario can be observed in Fig.…”

Section: A Implementation Details and Hyperparameter Valuesmentioning

confidence: 99%

“…Challenges. Despite growing interest towards analyzing memes, identifying the underlying connotations for the entities framed therein remains a challenging task (Sharma et al, 2022b). Memes are obscure due to their highly cryptic semantics and satirical content (Sabat et al, 2019).…”

Section: Introductionmentioning

confidence: 99%

“…Sharma et al (2022b) reannotate the HarMeme dataset(Pramanick et al, 2021b) by collecting annotator responses for different roles different entities within memes take. The re-annotated memes may or may not have harmful implications, contrary to distinction modelled as part of original curation.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Characterizing the Entities in Harmful Memes: Who is the Hero, the Villain, the Victim?

Sharma¹,

Kulkarni²,

Suresh³

et al. 2023

Preprint

View full text Add to dashboard Cite

Memes can sway people's opinions over social media as they combine visual and textual information in an easy-to-consume manner. Since memes instantly turn viral, it becomes crucial to infer their intent and potentially associated harmfulness to take timely measures as needed. A common problem associated with meme comprehension lies in detecting the entities referenced and characterizing the role of each of these entities. Here, we aim to understand whether the meme glorifies, vilifies, or victimizes each entity it refers to. To this end, we address the task of role identification of entities in harmful memes, i.e., detecting who is the 'hero', the 'villain', and the 'victim' in the meme, if any. We utilize HVVMemes -a memes dataset on US Politics and Covid-19 memes, released recently as part of the CONSTRAINT@ACL-2022 shared-task. It contains memes, entities referenced, and their associated roles: hero, villain, victim, and other. We further design VECTOR (Visual-semantic role dEteC-ToR), a robust multi-modal framework for the task, which integrates entity-based contextual information in the multi-modal representation and compare it to several standard unimodal (text-only or image-only) or multi-modal (im-age+text) models. Our experimental results show that our proposed model achieves an improvement of 4% over the best baseline and 1% over the best competing stand-alone submission from the shared-task. Besides divulging an extensive experimental setup with comparative analyses, we finally highlight the challenges encountered in addressing the complex task of semantic role labeling within memes.

show abstract

Section: Discussionmentioning

confidence: 97%