Elamur Research⌕
☰
home / papers / 2020 / Multimodal Meme Dataset (MultiOFF) for Identifying Offensive Content in Image and Text

Multimodal Meme Dataset (MultiOFF) for Identifying Offensive Content in Image and Text

Summary

Memes pair images with overlaid text and often carry offence implicitly, through sarcasm or a mismatch between picture and caption. A classifier that reads only the text or only the image can therefore miss the harm. Because no public multimodal resource for offensive meme detection existed, the authors built MultiOFF. They started from a Kaggle collection of memes about the 2016 U.S. presidential election, cleaned it, and had volunteer annotators label each meme as offensive or not-offensive. Their working definition of an offensive meme is one that aims to damage the social identity or lower the prestige of a person or community, a framing drawn from the SemEval 2019 offensive-language task. They compared text-only models, an image-only VGG16 model, and early-fusion models that concatenate text and image features. Adding visual features mainly increased recall for the neural text models. Overall scores remained modest, which shows how hard the task is. The dataset and code are public, giving a starting point for multimodal abuse detection.

Key points

  • MultiOFF dataset: 743 memes with binary labels (offensive / not-offensive), split randomly into train, validation and test sets. The memes originally come from Reddit, Facebook, Twitter and Instagram.
  • Preprocessing: Only the image URL and caption were kept from the Kaggle source. Noise symbols such as //n and @ were stripped, and posts with more than 20 sentences were dropped. Every image link was checked and the image downloaded locally.
  • Text-only baselines:
    • Logistic regression gave the best text-based results.
    • The text CNN reached the highest recall, but its precision was only 0.39.
    • DNN, which uses locally trained embeddings, reached an F1 of about 0.5 with recall 0.55.
  • Image-only baseline: VGG16 recalled just 0.16 of the offensive memes.
  • Multimodal models: Early fusion paired Stacked LSTM, BiLSTM or CNN text encoders with VGG16 image features.
    • All three averaged about 0.40 precision.
    • Recall improved over their text-only versions; for example, Stacked LSTM + VGG16 reached 0.66.

Interesting ideas

  • I1. Event-anchored data collection: Building the dataset around a single polarising event (an election) concentrates potentially offensive material. This could help bootstrap abuse datasets for other topics or languages. The authors also note the downside: domain bias that may hurt generalisation.
  • I2. Annotation aids for implicit content: Annotators disagreed on ambiguous memes, so they were given curated examples of offensive and non-offensive memes. A small gallery of reference cases is a cheap way to align judgments on subtle categories.
  • I3. Early fusion as a simple baseline: Concatenating features from a pretrained CNN (VGG16) with a text encoder gives a low-resource multimodal model that needs no object detectors. It is a sensible first step before trying more complex fusion of embeddings.
  • I4. Reporting recall and precision separately: Looking beyond F1 exposed trade-offs, such as the text CNN catching many offensive memes while mislabelling others. For moderation tasks, recall on the harmful class can matter more than a single aggregate score.
  • I5. Extensions proposed by the authors: Using OCR to pull embedded text out of images, and treating social media tags as weak labels. Both are practical ways to scale meme datasets without full manual annotation.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Natural language processing

Cited by

97 papers of our library cite it (of the 140884 whose bibliographies we read); 1 of them have a page here, 96 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{suryawanshi2020multimodal,
  title = {{Multimodal Meme Dataset (MultiOFF) for Identifying Offensive Content in Image and Text}},
  author = {Suryawanshi, Shardul and Chakravarthi, Bharathi Raja and Arcan, Mihael and Buitelaar, Paul},
  booktitle = {Proceedings of the Second Workshop on Trolling, Aggression and Cyberbullying},
  year = {2020},
  doi = {10.13025/21394},
  url = {https://doi.org/10.13025/21394}
}

References

23 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Arentz, W. A. and Olstad, B. (2004). Classifying offensive sites based on image content. Comput. Vis. Image Underst., 94(1-3):295–310, April. find
  2. Aroyehun, S. T. and Gelbukh, A. (2018). Aggression detection in social media: Using deep neural networks, data augmentation, and pseudo labeling. In Proceedings of the First Workshop on Trolling, Aggression and Cyberbullying (TRAC-2018), pages 90–97, Santa Fe, New Mexico, USA, August. Association fo… find
  3. Chakravarthi, B. R., Priyadharshini, R., Stearns, B., Jayapal, A., S, S., Arcan, M., Zarrouk, M., and McCrae, J. P. (2019). Multilingual multimodal machine translation for Dravidian languages utilizing phonetic transcription. In Proceedings of the 2nd Workshop on Technologies for MT of Low Resource… find
  4. Chakravarthi, B. R., Jose, N., Suryawanshi, S., Sherly, E., and McCrae, J. P. (2020a). A sentiment analysis dataset for code-mixed Malayalam-English. In Proceedings of the 1st Joint Workshop of SLTU (Spoken Language Technologies for Under-resourced languages) and CCURL (Collaboration and Computing f… find our card
  5. Chakravarthi, B. R., Muralidaran, V., Priyadharshini, R., and McCrae, J. P. (2020b). Corpus creation for sentiment analysis in code-mixed Tamil-English text. In Proceedings of the 1st Joint Workshop of SLTU (Spoken Language Technologies for Under-resourced languages) and CCURL (Collaboration and Com… find our card
  6. Jose, N., Chakravarthi, B. R., Suryawanshi, S., Sherly, E., and McCrae, J. P. (2020). A survey of current datasets for code-switching research. In 2020 6th International Conference on Advanced Computing & Communication Systems (ICACCS). find
  7. Kakumanu, P., Makrogiannis, S., and Bourbakis, N. (2007). A survey of skin-color modeling and detection methods. Pattern recognition, 40(3):1106–1122. find
  8. McCallum, A., Nigam, K., et al. (1998). A comparison of event models for naive bayes text classification. In AAAI-98 workshop on learning for text categorization, volume 752, pages 41–48. find
  9. Citeseer. Mojica de la Vega, L. G. and Ng, V. (2018). Modeling trolling in social media conversations. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan, May. European Language Resources Association (ELRA). find
  10. Pennington, J., Socher, R., and Manning, C. (2014). Glove: Global vectors for word representation. In Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pages 1532– 1543, Doha, Qatar. find
  11. Rani, P., Suryawanshi, S., Goswami, K., Chakravarthi, B. R., Fransen, T., and McCrae, J. P. (2020). A comparative study of different state-of-the-art hate speech detection methods for Hindi-English code-mixed data. In Proceedings of the Second Workshop on Trolling, Aggression and Cyberbullying, Mars… find
  12. Ranjan, P., Raja, B., Priyadharshini, R., and Balabantaray, R. C. (2016). A comparative study on code-mixed data of Indian social media vs formal text. In 2016 2nd International Conference on Contemporary Computing and Informatics (IC3I), pages 608–611, Dec. find
  13. Schmidt, A. and Wiegand, M. (2017). A survey on hate speech detection using natural language processing. In Proceedings of the Fifth International Workshop on Natural Language Processing for Social Media, pages 1–10, Valencia, Spain, April. Association for Computational Linguistics. Sharma, C., Paka… find our card
  14. Simonyan, K. and Zisserman, A. (2014). Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556. find our card
  15. Smitha, E., Sendhilkumar, S., and Mahalaksmi, G. (2018). Meme classification using textual and visual features. In Computational Vision and Bio Inspired Computing, pages 1015–1031. find
  16. Springer. Suryawanshi, S., Chakravarthi, B. R., Verma, P., Arcan, M., McCrae, J. P., and Buitelaar, P. (2020). A dataset for troll classification of Tamil memes. In Proceedings of the 5th Workshop on Indian Language Data Resource and Evaluation (WILDRE-5), Marseille, France, May. European Language R… find
  17. Tian, C., Zhang, X., Wei, W., and Gao, X. (2018). Color pornographic image detection based on color-saliency preserved mixture deformable part model. Multimedia Tools and Applications, 77(6):6629–6645, Mar. find
  18. Wang, Z. (2017). Temporal-related convolutional-restricted-boltzmann-machine capable of learning relational order via reinforcement learning procedure. International Journal of Machine Learning and Computing, 7:1–8, 02. find
  19. Warner, W. and Hirschberg, J. (2012). Detecting hate speech on the world wide web. In Proceedings of the second workshop on language in social media, pages 19–26, Stroudsburg, PA, USA. Association for Computational Linguistics. find our card
  20. Waseem, Z., Davidson, T., Warmsley, D., and Weber, I. (2017). Understanding abuse: A typology of abusive language detection subtasks. arXiv preprint arXiv:1705.09899. find our card
  21. Watanabe, H., Bouazizi, M., and Ohtsuki, T. (2018). Hate speech on twitter: A pragmatic approach to collect hateful and offensive expressions and perform hate speech detection. IEEE Access, 6:13825–13835. find
  22. Wiegand, M., Siegel, M., and Ruppenhofer, J. (2018). Overview of the germeval 2018 shared task on the identification of offensive language. Proceedings of GermEval 2018, 14th Conference on Natural Language Processing (KONVENS 2018), Vienna, Austria – September 21, 2018, pages 1 – 10. Austrian Academ… find
  23. Zampieri, M., Malmasi, S., Nakov, P., Rosenthal, S., Farra, N., and Kumar, R. (2019). SemEval-2019 task 6: Identifying and categorizing offensive language in social media (OffensEval). In Proceedings of the 13th International Workshop on Semantic Evaluation, pages 75–86, Minneapolis, Minnesota, USA,… find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/suryawanshi-2020-multimodal-meme-dataset-multioff-identifying/