Elamur Research⌕
☰
home / papers / 2018 / Cross-Target Stance Classification with Self-Attention Networks

Cross-Target Stance Classification with Self-Attention Networks

Paper · 2018 · Chang Xu, Cecile Paris, Surya Nepal et al.

Summary

Stance classifiers are usually trained and tested on a single target. This means every new topic needs its own annotated data. Xu et al. (CSIRO Data61) study cross-target stance classification instead: a model trained on one target is reused for a different but thematically related target. They propose CrossNet. It encodes the target with a BiLSTM and conditions a second BiLSTM over the sentence on that target encoding. A shared self-attention layer then highlights recurring domain "aspects", such as reef damage in environmental debates, before an MLP predicts the stance.

The model is evaluated on four SemEval-2016 Task 6 targets grouped into two domains, Women's Rights and American Politics. It is also qualitatively tested on unlabeled tweets about an Australian mining project, using a Climate Change model. CrossNet beats the BiLSTM and BiCond baselines in most settings, and its advantage is larger when transferring across targets. The attention weights pick out domain-relevant and sentiment words. The work points to a way of bootstrapping stance detection for new topics without fresh annotation.

Key points

  • Data: SemEval-2016 Task 6 targets CC, FM, HC, LA and DT, labelled favor, against or neither. A new AM corpus was also built: 220,067 tweets from January 2016 to June 2017, reduced to 40,852 unlabeled tweets after removing duplicates and URL-only posts.
  • Metric and setup: the score is the mean of micro- and macro-F1, and a transfer ratio compares cross-target with in-target performance. The model uses frozen 200d GloVe Twitter vectors and an LSTM hidden size of 60.
  • Cross-target F1: CrossNet scores 44.2 (FM→LA), 43.1 (LA→FM), 46.1 (HC→DT) and 41.8 (DT→HC). Its transfer ratios range from 0.74 to 0.83.
  • Gains over BiCond: the average improvement in F1 is 6.6% cross-target versus 3.0% in-target. Performance still drops by about 26% on average from the in-target to the cross-target setting.
  • Learned aspects: examples include "equality" and "feminist" for women's rights, "vote" and "government" for politics, and "reef", "destroy" and "environment" for the mining project.

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

🔬Data61Sydney · Australia17 works · 2,698 citations🏛️Commonwealth Scientific and Industrial Research OrganisationCanberra · Australia17 works · 2,064 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 works

Cite

Cite as Chang Xu, Cecile Paris, Surya Nepal and Ross Sparks. 2018. Cross-Target Stance Classification with Self-Attention Networks. arXiv:1805.06593.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.1805.06593 · arXiv 1805.06593

@misc{xu2018cross,
  title = {{Cross-Target Stance Classification with Self-Attention Networks}},
  author = {Chang Xu, Cecile Paris, Surya Nepal and Ross Sparks},
  year = {2018},
  doi = {10.48550/arxiv.1805.06593},
  eprint = {1805.06593},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1805.06593}
}

References

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

  1. Isabelle Augenstein, Tim Rocktäschel, Andreas Vlachos, and Kalina Bontcheva. 2016. Stance detection with bidirectional conditional encoding. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP). pages 876–885.
  2. Jianpeng Cheng, Li Dong, and Mirella Lapata. 2016. Long short-term memory-networks for machine reading. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP). pages 551–561.
  3. Javid Ebrahimi, Dejing Dou, and Daniel Lowd. 2016. Weakly supervised tweet stance classification by relational bootstrapping. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing. pages 1012–1017. find
  4. Xavier Glorot, Antoine Bordes, and Yoshua Bengio. 2011. Domain adaptation for large-scale sentiment classification: A deep learning approach. In Proceedings of the 28th international conference on machine learning (ICML). pages 513–520. find
  5. Kazi Saidul Hasan and Vincent Ng. 2013. Stance classification of ideological debates: Data, models, features, and constraints. In Proceedings of the 6th International Joint Conference on Natural Language Processing. pages 1348–1356.
  6. Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural Computation 9(8):1735–1780. find
  7. Diederik Kingma and Jimmy Ba. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 .
  8. Christopher D Manning, Prabhakar Raghavan, Hinrich Schütze, et al. 2008. Introduction to information retrieval, volume 1. Cambridge university press Cambridge. find
  9. Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013. Distributed representations of words and phrases and their compositionality. In Advances in neural information processing systems (NIPS). pages 3111–3119.
  10. Saif Mohammad, Svetlana Kiritchenko, Parinaz Sobhani, Xiao-Dan Zhu, and Colin Cherry. 2016. Semeval-2016 task 6: Detecting stance in tweets. In SemEval@ NAACL-HLT. pages 31–41.
  11. Saif M Mohammad, Parinaz Sobhani, and Svetlana Kiritchenko. 2017. Stance and sentiment in tweets. ACM Transactions on Internet Technology (TOIT) 17(3):26.
  12. Jeffrey Pennington, Richard Socher, and Christopher D. Manning. 2014. Glove: Global vectors for word representation. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP). pages 1532–1543.
  13. Mariona Taulé, M Antonia Martı́, Francisco M Rangel, Paolo Rosso, Cristina Bosco, Viviana Patti, et al. 2017. Overview of the task on stance and gender detection in tweets on catalan independence at ibereval 2017. In 2nd Workshop on Evaluation of Human Language Technologies for Iberian Languages, I… find
  14. 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 Advances in neural information processing systems (NIPS). pages 6000–6010.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xu-2018-cross-target-stance-classification-self/