Elamur Research⌕
☰
home / papers / 2022 / #maskUp: Selective Attribute Encryption for Sensitive Vocalization for English language on Social Media Platforms

#maskUp: Selective Attribute Encryption for Sensitive Vocalization for English language on Social Media Platforms

Paper · 2022 · Supriti Vijay and Aman Priyanshu

Summary

The paper addresses a conflict faced by people who report gender-based crimes on social media. Sharing details can help investigators, but publicly exposed names, places or organisations can lead to harassment of the victim. The authors propose #maskUp, a pipeline that leaves most of a post readable and encrypts only the spans that could identify people.

The pipeline has three parts:

  • Entity detection: a transformer NER model, trained through the NERDA framework on CoNLL-2003, finds identifying spans.
  • Per-user encryption: each user gets an AES key derived from their platform password, and only the detected entities are encrypted with it.
  • Authority access: these user keys are wrapped with RSA using a public key held by a legal authority, which can then unmask any account.

The NER component is meant to keep learning from user corrections, using Elastic Weight Consolidation to limit forgetting. The evaluation covers only small samples. Selective encryption ran roughly ten times faster than encrypting whole texts and used less memory. The authors argue this makes privacy-preserving reporting cheap enough for edge devices.

Key points

  • NER uses "google/electra-small-discriminator" via the NERDA library. It is trained on CoNLL-2003 with BIO tags for PER, ORG, LOC and MISC.
  • 5-fold cross-validation gives two averaged F1 values, 0.894 and 0.853. Per-tag F1 ranges from 0.984 (I-PER) and 0.950 (B-PER) down to 0.645 (I-MISC).
  • The per-user key is 128 bits. It is built from the platform password through a noise transformation followed by double encryption.
  • Averaged over 30 articles, full AES took 2038.5 ms to encrypt and 1970.4 ms to decrypt. #maskUp took 211.6 ms and 207.5 ms.
  • Memory use fell from 1.406 kB to 0.553 kB. The authors report gains of up to 89.1% in encryption time and 60.7% in memory.
  • Continual learning follows EWC (Fisher-weighted penalty on parameter drift), building on the authors' NERDA-Con library.

Interesting ideas

  • I1. Transform only the identifying spans and leave the surrounding text untouched, so the content stays useful while exposure drops.
  • I2. Treat user corrections of what counts as sensitive as a stream of training signal, and apply a Fisher-weighted EWC penalty so old skills are not overwritten.
  • I3. Measure the selective method against a "process everything" baseline in both time and memory, so the benefit is concrete rather than assumed.
  • I4. Two layers of keys separate everyday access (each user's own key) from oversight (an authority's master key that unwraps the user keys).
  • I5. Low MISC scores show that broad catch-all entity categories are where sensitive-span detection is weakest.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (back): the text → graph → text compressor could replace entity nodes with reversible placeholders, producing anonymised retellings with the same event structure.
  • I2 (learn): when retraining the perceptron parser or gparse on new silver data, an EWC-style term could protect accuracy on earlier treebanks, checked with held-out negative controls.
  • I3 (learn): explainable stages could be benchmarked against a full-pipeline baseline whenever a shortcut (e.g., parsing only relevant sentences) is introduced.
  • I4 (world): the world of a text could tag entity facts with visibility levels, so answers reveal identities only to readers with the right permission.
  • I5 (frames): identifying who is a person, place or organisation could rely on semantic roles and coreference chains rather than NER labels alone, catching identifiers that a MISC tag misses.

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

 Institutions

🎓Manipal Academy of Higher EducationManipal · India2 works · 17 citations

📚 Shelves 4 Computer Science · 9,086 works Artificial Intelligence · 8,889 works Social Sciences · 744 works Sociology and Political Science · 221 works

🧭 Topics 1 English · 1,373 works

Cite

Cite as Supriti Vijay and Aman Priyanshu. 2022. #maskUp: Selective Attribute Encryption for Sensitive Vocalization for English language on Social Media Platforms. arXiv:2211.08653.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2211.08653 · arXiv 2211.08653

@misc{vijay2022maskup,
  title = {{\#maskUp: Selective Attribute Encryption for Sensitive Vocalization for English language on Social Media Platforms}},
  author = {Supriti Vijay and Aman Priyanshu},
  year = {2022},
  doi = {10.48550/arxiv.2211.08653},
  eprint = {2211.08653},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2211.08653}
}

References

30 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. ???? WHO 2021: Violence against Women data. https://www.who.int/health-topics/violence-against-women. ↗
  2. Bender, O.; Och, F. J.; and Ney, H. 2003. Maximum Entropy Models for Named Entity Recognition. In Proceedings of the Seventh Conference on Natural Language Learning at HLT-NAACL 2003 - Volume 4, CONLL ’03, 148–151. USA: Association for Computational Linguistics. find
  3. Caruana, R. 1997. Multitask Learning. Machine Learning, 28: 1573–0565. find
  4. Chieu, H. L.; and Ng, H. T. 2002. Named Entity Recognition: A Maximum Entropy Approach Using Global Information. In COLING 2002: The 19th International Conference on Computational Linguistics. find
  5. Chiu, J. P. C.; and Nichols, E. 2015. Named Entity Recognition with Bidirectional LSTM-CNNs. ↗ our card
  6. Cucchiarelli, A.; and Velardi, P. 2001. Unsupervised Named Entity Recognition Using Syntactic and Semantic Contextual Evidence. Computational Linguistics, 27(1): 123–131. ↗
  7. Durrett, G.; and Klein, D. 2014. A Joint Model for Entity Analysis: Coreference, Typing, and Linking. Transactions of the Association for Computational Linguistics, 2: 477– 490. ↗ our card
  8. Etaiwi, W. A.; and Hraiz, S. 2018. Structured encryption algorithm for text cryptography. Journal of Discrete Mathematical Sciences and Cryptography, 21(7-8): 1559–1572. ↗
  9. Ghaddar, A.; Langlais, P.; Rashid, A.; and Rezagholizadeh, M. 2021. Context-aware Adversarial Training for Name Regularity Bias in Named Entity Recognition. Transactions of the Association for Computational Linguistics, 9: 586– 604. ↗
  10. Hovy, D.; and Spruit, S. L. 2016. The Social Impact of Natural Language Processing. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), 591–598. Berlin, Germany: Association for Computational Linguistics. our card
  11. Kirkpatrick, J.; Pascanu, R.; Rabinowitz, N.; Veness, J.; Desjardins, G.; Rusu, A. A.; Milan, K.; Quan, J.; Ramalho, T.; Grabska-Barwinska, A.; Hassabis, D.; Clopath, C.; Kumaran, D.; and Hadsell, R. 2016. Overcoming catastrophic forgetting in neural networks. ↗
  12. Kishor, S.; and Johnson, K. 2004. Profiling domestic violence. A multy - country study. find
  13. Koss, M. P. 1993. Rape. Scope, impact, interventions, and public policy responses. The American psychologist, 48 10: 1062–9. find
  14. Kushwaha, A.; Sharma, H. R.; and Ambhaikar, A. 2016. A Novel Selective Encryption Method for Securing Text Over Mobile Ad Hoc Network. Procedia Computer Science, 79: 16–23. Proceedings of International Conference on Communication, Computing and Virtualization (ICCCV) 2016. find
  15. Kushwaha, A.; Sharma, H. R.; and Ambhaikar, A. 2018. Selective Encryption Using Natural Language Processing for Text Data in Mobile Ad Hoc Network, 15–26. Cham: Springer International Publishing. ISBN 978-3-319-70542-2. find
  16. Lample, G.; Ballesteros, M.; Subramanian, S.; Kawakami, K.; and Dyer, C. 2016. Neural Architectures for Named Entity Recognition. ↗ our card
  17. Li, Z.; and Hoiem, D. 2017. Learning without Forgetting. arXiv:1606.09282. ↗
  18. Mishra, S.; He, S.; and Belli, L. 2020. Assessing Demographic Bias in Named Entity Recognition. ↗
  19. Nadeau, D.; Turney, P. D.; and Matwin, S. 2006. Unsupervised Named-Entity Recognition: Generating Gazetteers and Resolving Ambiguity. In Proceedings of the 19th International Conference on Advances in Artificial Intelligence: Canadian Society for Computational Studies of Intelligence, AI’06, 266–277. Berlin, Heidelberg: Springer-Verlag. ISBN 3540346287. find
  20. Parisi, G. I.; Kemker, R.; Part, J. L.; Kanan, C.; and Wermter, S. 2019. Continual lifelong learning with neural networks: A review. Neural Networks, 113: 54–71. ↗ our card
  21. Pasca, M.; Lin, D.; Bigham, J.; Lifchits, A.; and Jain, A. 2006. Organizing and Searching the World Wide Web of Facts - Step One: The One-Million Fact Extraction Challenge. In Proceedings of the 21st National Conference on Artificial Intelligence - Volume 2, AAAI’06, 1400–1405. AAAI Press. ISBN 9781577352815. find
  22. Riloff, E.; and Jones, R. 1999. Learning Dictionaries for Information Extraction by Multi-Level Bootstrapping. In Proceedings of the Sixteenth National Conference on Artificial Intelligence and the Eleventh Innovative Applications of Artificial Intelligence Conference Innovative Applications of Artificial Intelligence, AAAI ’99/IAAI ’99, 474–479. USA: American Association for Artificial Intelligence. ISBN 0262511061. find
  23. Roy, A. 2021. Recent Trends in Named Entity Recognition (NER). Santos, C. N. d.; and Guimarães, V. 2015. Boosting Named Entity Recognition with Neural Character Embeddings. ↗ our card
  24. Settles, B. 2004. Biomedical Named Entity Recognition Using Conditional Random Fields and Rich Feature Sets. In Proceedings of the International Joint Workshop on Natural Language Processing in Biomedicine and Its Applications, JNLPBA ’04, 104–107. USA: Association for Computational Linguistics. our card
  25. Suvarna, A.; and Bhalla, G. 2020. #NotAWhore! A Computational Linguistic Perspective of Rape Culture and Victimization on Social Media. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop, 328–335. Online: Association for Computational Linguistics. find
  26. Vijay, S.; and Priyanshu, A. 2022. NERDA-Con: Extending NER models for Continual Learning – Integrating Distinct Tasks and Updating Distribution Shifts. ↗
  27. Yadav, V.; and Bethard, S. 2019. A Survey on Recent Advances in Named Entity Recognition from Deep Learning models. ↗ our card
  28. Yan, H.; Deng, B.; Li, X.; and Qiu, X. 2019. TENER: Adapting Transformer Encoder for Named Entity Recognition. ↗
  29. Yangarber, R.; Lin, W.; and Grishman, R. 2002. Unsupervised Learning of Generalized Names. In Proceedings of the 19th International Conference on Computational Linguistics - Volume 1, COLING ’02, 1–7. USA: Association for Computational Linguistics. find
  30. Zhou, G.; and Su, J. 2002. Named Entity Recognition using an HMM-based Chunk Tagger. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, 473–480. Philadelphia, Pennsylvania, USA: Association for Computational Linguistics. *Figure 2: Algorithmic flow of a User’s post through #maskUp.* our card

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/vijay-2022-maskup-selective-attribute-encryption-sensitive/