Victoria Sherratt — memes, misinformation and multimodal meaning

Research

Internet memes generate meaning through the interaction of their visual and text modalities, interpreted alongside cultural knowledge external to the meme itself. Format and meaning are often decoupled; visually similar memes carry distinct cultural meanings, whilst semantically related memes may share no perceptual similarity at all. Systems that treat memes as plain image-text pairs without cultural context fail in the critical cases like harmful, misinformative and propagandistic content.

My research moves the machine reading of internet culture from pattern matching to cultural grounding; it treats meaning as something communities construct, encode and change, and builds the semiotic and computational instruments to observe meaning at scale. My primary area of focus is internet memes and online communications.

My doctoral work and research addressed this with SemioMeme, a knowledge graph built on the principle that meme meaning and appearance are separable representations.

Most recently I have formalised community-authored usage glosses as use-conditions for meme formats, using informal community documentation to predict measurable regularities in meme production across linguistic and semantic content, alongside the function of memes as communicative devices.

My research also deals with the open problem of actualisation in internet memes, symbols and coded communication; whilst we produce many catalogues to track symbols or icons, including harmful content, few resources categorise the most severe cases of coded derivatives that rapidly emerge through users repurposing materials and evading moderation detection. Meaning outruns curated records; current approaches match only known symbols, and systems trained on this data fail in the most crucial cases like coded communication, dog-whistles or emerging harm. My current work approaches this as a question of dynamics; how a second, coded meaning attaches to a symbol, spreads and settles into convention.

decoupledthe same image can be benign in one community and coded in another, with nothing in the image changing; and one meaning can travel across formats that share no visual features. What a meme looks like and what it means are separate facts.

SemioMemesemiotics, meme; a graph in which a meme's cultural relations, to people, events, subcultures and other memes, are recorded alongside its visual and textual form, so either can be queried against the other.

use-conditionsthe meaning of an expression given by when it is appropriately used rather than what it denotes, which can be extracted from informal documentation of internet memes as an additional organising principle to semantic similarity.

actualisationa symbol's meaning is potential until it is used; the same frog is actualised as a joke in one thread and a call-sign in another.

dynamicssemiotics has a rich theory of structure and an anecdotal theory of change, whereas computational analysis detects change but cannot say what it means. I work in this overlap.

Publications

Six first-author papers, 2022–2026. Venues: ICWSM, ICMR, ACL, IJCAI, SemDial.

Sherratt, V., Dethlefs, N. (2026). Community-Authored Usage Glosses as Use-Conditions for Internet Memes. Proceedings of the 30th Workshop on Semantics and Pragmatics of Dialogue. Accepted.
Sherratt, V., Elayan, S. & Dethlefs, N. (2026). SemioMeme: A Symbolic–Subsymbolic Knowledge Graph Dataset for Multimodal Meme Analysis. Proceedings of the International AAAI Conference on Web and Social Media, 20(1), 2921–2935. Paper
Sherratt, V., Dethlefs, N. (2026). Clustering Internet Memes with Metric Learning and Dynamic Modality Weighting. Proceedings of the International AAAI Conference on Web and Social Media, 20(1), 2131–2148. Paper
Sherratt, V., Dogan, S., Wuraola, I., Bryan-Smith, L., Onwuchekwa, O., & Dethlefs, N. (2024). BDA at SemEval-2024 Task 4: Detection of Persuasion in Memes Across Languages with Ensemble Learning and External Knowledge. Proceedings of SemEval-2024, 123–132. ACL. Paper
Sherratt, V., Pimbblet, K., & Dethlefs, N. (2023). Multi-channel Convolutional Neural Network for Precise Meme Classification. Proceedings of ICMR 2023, 190–198. ACM. Paper
Sherratt, V. (2022). Towards Contextually Sensitive Analysis of Memes: Meme Genealogy and Knowledge Base. Proceedings of IJCAI-22 Doctoral Consortium, 5871–5872. Paper

2026, glossesmakes community-documented function machine-readable; a format's stated use-conditions predict how it is produced and its communicative function.

2026, SemioMemeargues meme meaning and appearance need separate representations; the graph makes cultural association queryable alongside perceptual similarity.

2026, clusteringno single modality carries meme similarity; weighting modalities per query outperforms unified multimodal models at aligning memes to knowledge sources.

2024named visual entities as language-independent external knowledge; the ensemble generalised to languages it never saw natively.

2023argues the field's datasets mistake adverts, posters and screenshots for memes, and that live evaluation, not benchmarks, is the real test.

2022the doctoral consortium paper; proposed the programme, genealogy to knowledge base to use-conditions, that the later papers carried out.

Selected Projects

SemioMeme

ICWSM 2026 · doi.org/10.5281/zenodo.17826799

A symbolic–subsymbolic multimodal knowledge graph and dataset resource for studying internet meme culture as a connected ecosystem. An RDF knowledge graph of cultural relationships between meme concepts is combined with multimodal embeddings for perceptual similarity at instance level, making the cultural context embedded in memes computationally accessible alongside meme form. The complete resource is openly available.

Meme clustering at scale

ICWSM 2026 · Code

A multimodal metric learning approach for clustering internet memes into existing knowledge bases at scale, over 678,000 memes in our experiments, with dynamic modality weighting to balance image and text per meme.

SemEval-2024 Task 4, multilingual persuasion detection

Team lead, BDA (University of Hull) · 2024

Put together and led a five-person research team for the shared task. The system combined a multimodal ensemble with named visual entities as external knowledge, joined by late fusion; the system paper was published at ACL.

Alan Turing Institute Enrichment Scheme

Community Award · 2024

A nine-month placement within the Turing's cross-disciplinary research community, with development funding.

Agentic pipeline for higher-education course data

2020–2026

An annual data cycle covering course collection, merging, subject recoding and validation across UK universities, redesigned as an agent-run workflow with staged execution and human confirmation gates. Core elements include the Model Context Protocol (MCP), tool function calling, task decomposition, multi-agent orchestration, LLM-as-judge evaluation and human-in-the-loop design.

Experience

Co-founder & Director of Data Science

S Squared Insights Limited · 2020–2026

Co-founded a data analytics consultancy serving over 40 higher-education institutions. Responsible for the company's technical work: analytical methodology, infrastructure, data strategy and governance. Recruited, mentored and led the data science team. Shipped production systems including agentic AI pipelines, pattern-recognition tools and custom analytical products.

Research Data Analyst / Student Insight Officer

University of Manchester · 2019–2021

Lead for data science projects in the Market and Student Insight Team, and coordinator for the introduction of machine learning to the team's remit. Predictive models for student attainment and matriculation, market analysis for new course development, and analytics products for research and REF strategy.

Researcher

Fairhome Group PLC · 2018–2019

Data-driven business intelligence and research briefs for external stakeholders. Lead for GIS rollout across the business.

Education

PhD Computer Science, Loughborough University, 2026
MSc Artificial Intelligence & Data Science, University of Hull, Distinction, 2021
MA Contemporary Literature & Critical Theory, Manchester Metropolitan University, Distinction, 2018
BA (Hons) English & Creative Writing, Manchester Metropolitan University, First Class, 2015

Teaching

Lab Demonstrator, Loughborough University, 2026. Foundations for Artificial Intelligence modules, plus guest lectures.
Graduate Teaching Assistant, University of Hull, 2021–2023. Programming for Data Science and AI; Fundamentals of Data Science; Understanding Artificial Intelligence; Big Data and Data Mining; Applied Artificial Intelligence.

Awards

Alan Turing Institute Enrichment Scheme, Community Award, 2024.
Office for Students Artificial Intelligence Scholarship, 2017.
Departmental Award for Outstanding Academic Achievement, 2016.
Master Scholar, 2016.