SemioMeme: A Symbolic–Subsymbolic Knowledge Graph Dataset for Multimodal Meme Analysis
收藏资源简介：
If you use this dataset, please cite: 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. https://doi.org/10.1609/icwsm.v20i1.42792 BibTeX: @article{Sherratt_Elayan_Dethlefs_2026, title = {SemioMeme: A Symbolic--Subsymbolic Knowledge Graph Dataset for Multimodal Meme Analysis}, author = {Sherratt, Victoria and Elayan, Suzanne and Dethlefs, Nina}, journal = {Proceedings of the International AAAI Conference on Web and Social Media}, volume = {20}, number = {1}, pages = {2921--2935}, year = {2026}, month = may, doi = {10.1609/icwsm.v20i1.42792}, url = {https://ojs.aaai.org/index.php/ICWSM/article/view/42792} } SemioMeme is a multi-layer knowledge graph and dataset resource for internet meme analysis, comprising: Meta Layer: RDF knowledge graph encoding cultural relationships between meme concepts, people, events, sites, and subcultures from KnowYourMeme.com Corpus Layer: Extended graph (7.4M triples) linking meme instances to semantic context in the Meta Layer Embedding Layer: FAISS-indexed vision (SigLIP) and text (SentenceBERT) embeddings with explicit bridges to symbolic representations Also included: Source data (KYM metadata, image URLs, OCR-extracted text) Raw and fine-tuned embeddings Fine-tuned retrieval models Complete construction pipelines (GitHub code) Available on request from correspdoning author (v.sherratt@lboro.ac.uk): Raw images used in construction (~180GB depending on required subsets) For full documentation, see the accompanying paper, README.md and DATASHEET.md Paper: https://doi.org/10.1609/icwsm.v20i1.42792Code: GitHub Image Availability: Due to UK legislation and hosting restrictions, raw images are available upon request only. We appreciate that relying on URLs impacts the reproducibility and longevity of SemioMeme; users of the resource are strongly encouraged to contact the corresponding author (Victoria Sherratt, v.sherratt@lboro.ac.uk), who is happy to support access and reproducibility efforts where possible.
