What Makes a Meme a Meme? - Identifying Memes for Memetics-Aware Dataset Creation

Aug 26, 2026

Upload by Maeve

Multimodal Internet memes shape online discourse, yet meme classification often relies on supervised models trained on static datasets that ignore memetics—the imitation and transformation process defining genuine memes. This work proposes a meme identification protocol that detects memetic elements to distinguish memes from non-memetic content. Applied to a random sample of seven leading meme classification datasets, the protocol finds over half (50.4%) contain no memetic signs. The study also provides a memetics-grounded meme typology to improve interpretation and dataset creation.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

AccessFree to download

Abstract

1 Introduction

What Makes a Meme a Meme? Identifying Memes for Memetics-Aware Dataset Creation Muzhaffar Hazman 1 , Susan McKeever2 , Josephine Griffith 1 1University of Galway, Galway, Ireland. 2TU Dublin, Dublin, Ireland. [m.hazman1@universityofgalway.ie](m.hazman1@universityofgalway.ie), [susan.mckeever@tudublin.ie](susan.mckeever@tudublin.ie), [josephine.griffith@universityofgalway.ie](josephine.griffith@universityofgalway.ie) Abstract WARNING: This paper contains memes that may be offensive to some readers. Multimodal Internet Memes are now a ubiquitous fixture in online discourse. One strand of meme-based research is the classification of memes according to various affects, such as sentiment and hate, supported by manually compiled meme datasets. Understanding the unique characteristics of memes is crucial for meme classification. Unlike other user-generated content, memes spread via memetics, i.e. the process by which memes are imitated and transformed into elements used to create new memes. In effect, there exists an ever-evolving pool of visual and linguistic elements that underpin meme culture and are crucial to interpreting the meaning of individual memes. The current approach of training supervised learning modelson static datasets, without taking memetics into account, limits the depth and accuracy of meme interpretation. We argue that meme datasets must contain genuine memes, as defined via memetics, so that effective meme classifiers can be built. In this work, we develop a meme identification protocol which distinguishes meme from non-memetic content by recognising the memetic element within it. We apply our protocol to a random sample of the leading 7 meme classification datasets and observe that more than half (50 . 4%) of the evaluated samples were found to contain no signs of memetics. Our work also provides a meme typology grounded in memetics, providing the basis for more effective approaches to the interpretation of memes and the creation of meme datasets. 1 Introduction Internet Memes have become a staple of expression and interaction within digital communities. Although commonly recognised as multimodal humourous jokes, memes present a medium to participate in a global digital culture (Shifman 2014b), spread ideologies (Zannettou et al. 2018), debate political issues (Shifman 2014b), and even distribute propaganda (Dimitrov et al. 2021) . The expressive nature of memes has led to recent interest in classifying them by the various affects they convey including sentiment, hate, humour, sarcasm and offensiveness. The goals underlying meme classification are varied, e.g. identifying hateful memes to help kerb the proliferation of hateful rhetoric (Sharma et al. 2022), detecting signs of cyberbullying (Shang et al. 2021b), recognising Copyright © 2025, Association for the Advancement of Artificial Intelligence ([www.aaai.org](www.aaai.org)). All rights reserved. trolling behaviour (Suryawanshi et al. 2020), and discerning the sentiment conveyed by memes (Sharma et al. 2020) . Meme classification datasets provide memes labelled with various affects to aid in the development of meme classifiers via supervised learning. The current state-of-the-art in collecting and labelling sample memes is almost a completely manual process. Since not all content on the Internet are memes, the authors of these datasets have indicated the need to identify memes from non-memetic content during data collection. However, none of the 12 publicly available peerreviewed meme classification datasets we surveyed disclosed any standards or criteria used to identify memes from nonmemetic content. For clarity, we use meme classification to describe the task of classifying the various affects conveyed by memes as posed by these datasets, while we refer to meme identification as the task of selecting memes from non-memetic content (non-memes) . Work in both the fields of computational and philological research on memes identi

FAQs

Why can current meme classification approaches be limited?

They often train supervised models on static datasets without accounting for memetics, which restricts the depth and accuracy of meme interpretation.

How does the proposed work identify whether content is a meme?

It introduces a meme identification protocol that distinguishes meme from non-memetic content by recognizing the memetic element inside the content.

What did the authors observe when applying the protocol to existing datasets?

In a random sample of seven leading meme classification datasets, more than half (50.4%) showed no signs of memetics.

Aug 26, 2026

Upload by Maeve