Elamur Research⌕
☰
home / papers / 2021 / Adversarial Learning for Zero-Shot Stance Detection on Social Media

Adversarial Learning for Zero-Shot Stance Detection on Social Media

Summary

This paper tackles zero-shot stance detection on Twitter, where a model must classify a tweet's stance (pro, con, neutral) toward a topic it never saw during training. Twitter stance datasets cover few topics with many examples each, so the authors frame topic transfer as domain adaptation and treat each topic as a separate domain. Their model, TOAD (TOpic-ADversarial Network), builds on BiCond conditional encoding with topic-guided attention. A topic discriminator trained through gradient reversal pushes the document representation to become topic-invariant. Reconstruction losses, a near-identity transformation layer and a residual link to the topic encoding keep stance cues from being lost. On SemT6, they train on five topics and test on the sixth, rotating through all six with no assumed link between training and test topics. TOAD reaches state-of-the-art results on several topics and needs less computation than pre-trained language models such as BERT. It also gives the first zero-shot benchmarks for two topics that earlier cross-target work had left out.

Key points

  • Data: The paper uses the SemEval2016 Task 6 (SemT6) dataset, which has six topics: DT, HC, FM, LA, CC and A. Each topic has 564–984 labeled tweets. The discriminator also uses about 2k unlabeled tweets per test topic from the Augenstein et al. (2016) collection, selected with 1–2 keywords.
  • Architecture: TOAD combines BiLSTM topic and text encoders with scaled dot-product attention and 100-d GloVe embeddings. A feed-forward stance classifier works alongside an adversarial topic discriminator.
  • Results (F avg):
    • TOAD scores 49.5 on DT, 51.2 on HC and 54.1 on FM, beating BERT, BiCond, CrossN, VTN and SEKT on these topics.
    • SEKT remains stronger on LA (53.6 vs. 46.2).
    • BERT remains stronger on A and CC.
  • Ablation: On DT, removing the adversary drops the score from 49.5 to 41.5. Removing the residual topic connection costs 10.2 points on DT and 7.8 on HC. Removing the unlabeled data costs 9.5 points on DT.
  • Analysis: KMeans clustering shows that TOAD's representations align less with topic labels than the model without the adversary.
  • Availability: Code is released at github.com/MalavikaSrikanth16/adversarial-learning-for-stance.

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: Linguistics & languages

Cited by

41 papers of our library cite it (of the 140673 whose bibliographies we read); 15 of them have a page here, 26 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2021.naacl-main.379 · arXiv 2105.06603

@inproceedings{allaway2021adversarial,
  title = {{Adversarial Learning for Zero-Shot Stance Detection on Social Media}},
  author = {Allaway, Emily and Srikanth, Malavika and McKeown, Kathleen},
  booktitle = {Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies},
  year = {2021},
  doi = {10.18653/v1/2021.naacl-main.379},
  eprint = {2105.06603},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.18653/v1/2021.naacl-main.379}
}

References

22 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Emily Allaway and Kathleen McKeown. 2020. Zero-shot stance detection: A dataset and model using generalized topic representations. In EMNLP. find our card
  2. Isabelle Augenstein, Tim Rocktäschel, Andreas Vlachos, and Kalina Bontcheva. 2016. Stance detection with bidirectional conditional encoding. In EMNLP. find our card
  3. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. ArXiv, abs/1810.04805. find
  4. Yaroslav Ganin and Victor Lempitsky. 2015. Unsupervised domain adaptation by backpropagation. In ICML. find
  5. S. Hochreiter and J. Schmidhuber. 1997. Long short-term memory. Neural Computation, 9:1735–1780. find
  6. Diederik P. Kingma and Jimmy Ba. 2015. A method for stochastic optimization. abs/1412.6980. find
  7. Adam: CoRR, Dilek Küçük. 2017. Stance detection in turkish tweets. ArXiv, abs/1706.06894. — [arXiv:1706.06894](https://arxiv.org/abs/1706.06894) find
  8. Mirko Lai, Alessandra Teresa Cignarella, D. I. H. Farías, Cristina Bosco, V. Patti, and P. Rosso. 2020. Multilingual stance detection in social media political debates. Comput. Speech Lang., 63:101075. find
  9. J. Lin. 1991. Divergence measures based on the shannon entropy. IEEE Trans. Inf. Theory, 37:145–151. find
  10. Saif M. Mohammad, Svetlana Kiritchenko, Parinaz Sobhani, Xiao-Dan Zhu, and Colin Cherry. 2016. Semeval-2016 task 6: Detecting stance in tweets. In SemEval@NAACL-HLT. find
  11. Jeffrey Pennington, Richard Socher, and Christopher D. Manning. 2014. Glove: Global vectors for word representation. In Empirical Methods in Natural Language Processing (EMNLP), pages 1532–1543. 2 https://developer.twitter.com/en/ developer-terms/agreement-and-policy Barbara Plank and Gertjan van No… ↗
  12. Sebastian Ruder and Barbara Plank. 2017. Learning to select data for transfer learning with bayesian optimization. In EMNLP. find
  13. Mariona Taulé, Maria Antònia Martí, Francisco M. Rangel Pardo, Paolo Rosso, Cristina Bosco, and Viviana Patti. 2017. Overview of the task on stance and gender detection in tweets on catalan independence. In IberEval@SEPLN. find
  14. Adam Tsakalidis, Nikolaos Aletras, Alexandra I. Cristea, and Maria Liakata. 2018. Nowcasting the stance of social media users in a sudden vote: The case of the greek referendum. Proceedings of the 27th ACM International Conference on Information and Knowledge Management. find
  15. Jannis Vamvas and Rico Sennrich. 2020. X -stance: A multilingual multi-target dataset for stance detection. In KONVENS. find
  16. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Proceedings of the Thirty-First Annual Conference on Neural Information Systems. find
  17. Penghui Wei and W. Mao. 2019. Modeling transferable topics for cross-target stance detection. Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval. find
  18. Wan Wei, Xiao Zhang, Xuqin Liu, Wei Chen, and Tengjiao Wang. 2016. pkudblab at semeval-2016 task 6 : A specific convolutional neural network system for effective stance detection. In SemEval@NAACL-HLT. find our card
  19. Chang Xu, Cécile Paris, Surya Nepal, and Ross Sparks. 2018. Cross-target stance classification with self-attention networks. In ACL. find
  20. Guido Zarrella and Amy Marsh. 2016. Mitre at semeval-2016 task 6: Transfer learning for stance detection. In SemEval@NAACL-HLT. find our card
  21. B. Zhang, M. Yang, Xutao Li, Yunming Ye, Xiaofei Xu, and Kuai Dai. 2020. Enhancing cross-target stance detection with transferable semantic-emotion knowledge. In ACL. find our card
  22. Yuan Zhang, Regina Barzilay, and Tommi S. Jaakkola. 2017. Aspect-augmented adversarial networks for domain adaptation. Transactions of the Association for Computational Linguistics, 5:515–528. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/allaway-2021-adversarial-learning-zero-shot-stance/