--- title: "Stance Classification of Context-Dependent Claims (2017)" description: "The paper introduces claim stance classification: given a controversial debate topic… 51.9% accuracy. 3 references, cited by 78 papers here." authors: ["Roy Bar-Haim", "Indrajit Bhattacharya", "Francesco Dinuzzo", "Amrita Saha", "Noam Slonim"] year: 2017 venue: "Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers" source: "https://aclanthology.org/E17-1024/" licence: "CC BY 4.0 (ACL Anthology)" licence_url: "https://creativecommons.org/licenses/by/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/haim-2017-stance-classification-context-dependent-claims/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/haim-2017-stance-classification-context-dependent-claims/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2017](https://md.research.elamur.ai/free/en/papers/2017/page/3/) / Stance Classification of Context-Dependent Claims # Stance Classification of Context-Dependent Claims **Authors:** [Roy Bar-Haim](https://md.research.elamur.ai/free/en/authors/roy-bar-haim/), Indrajit Bhattacharya, Francesco Dinuzzo, [Amrita Saha](https://md.research.elamur.ai/free/en/authors/amrita-saha/) and [Noam Slonim](https://md.research.elamur.ai/free/en/authors/noam-slonim/) · **Year:** 2017 · **Venue:** [Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers](https://md.research.elamur.ai/free/en/venues/eacl/) ([EACL 2017](https://md.research.elamur.ai/free/en/venues/eacl/2017/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/E17-1024/) · [PDF](https://aclanthology.org/E17-1024.pdf) **Citations:** [78](https://md.research.elamur.ai/free/en/papers/haim-2017-stance-classification-context-dependent-claims/cited-by/) in our library, of the 141,840 papers whose bibliographies we read ## Summary The paper introduces claim stance classification: given a controversial debate topic and claims already extracted for it, decide whether each claim argues for (Pro) or against (Con) the topic. To create the task's first benchmark, the authors annotate an extended version of the IBM argumentative structure dataset (Wikipedia claims for IDEA debate motions) with majority-vote Pro/Con labels. Instead of one end-to-end classifier, they split the decision into three parts. First, find the target phrase in both topic and claim. Second, determine the sentiment toward each target. Third, decide whether the two targets point the same way (consistent) or opposite ways (contrastive). Stance is the product of these three signs. A continuous version gives confidence scores, so the system can output only predictions it trusts. Annotation shows the decomposition fits about 95% of claims. The modular system, which includes a new contrast-detection method, beats unigram-based SVM baselines across coverage levels. This matters for open-domain settings, where topics are not known ahead of time and no topic-specific training data exists. ## Key points - **Dataset:** an updated IBM argumentative structure dataset with 55 topics and 2,394 claims. Five annotators labelled each claim for stance, and the majority label was kept. - **Model:** Stance(c, t) = s_c × R(x_c, x_t) × s_t. - s_c and s_t are the sentiments of the claim and the topic toward their targets. - R is +1 when the targets are consistent and −1 when they are contrastive. - **Coverage of the model:** annotation found that roughly 95% of claims fit this target-based decomposition, and contrastive targets turned out to be common. - **Baselines:** Unigrams SVM and Unigrams+Sentiment SVM. The majority baseline reaches 51.9% accuracy. - **Results by configuration:** - The full system with contrast detection reaches 83.6% accuracy at 30% coverage and 79.3% at 40%. - Combining the system with the unigram SVM gives the best accuracy at high coverage (≥0.8). ## Interesting ideas - **Sign-product decomposition:** reducing stance to three simpler sub-decisions makes errors easier to trace and lets each part be trained or annotated separately. The same idea could be reused for any relational polarity task. - **Accuracy-at-coverage evaluation:** the authors search over confidence thresholds to find the best accuracy for each minimum coverage level. This fits applications that show users only confident outputs, and it rewards good confidence ranking rather than just raw accuracy. - **Using the contrast score for re-ranking:** multiplying the stance score by the contrast classifier's output pushes claims with clearly consistent targets to the top. This shows how a secondary relation model can sharpen confidence without changing the main predictor. - **Hybrid fallback:** the precise but low-coverage semantic pipeline is paired with a broad lexical SVM so that claims the pipeline cannot handle still receive a prediction. This is a practical pattern for rule-heavy or resource-dependent components. - **Filtering anchor candidates:** candidate phrases are drawn from several sources (parser, Wikifier, NER, WordNet multiword expressions). Only terms that are statistically dominant for the topic are kept, using a hypergeometric test with Bonferroni correction. This cheaply focuses the analysis on topic-relevant terms. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* 📚 Shelves (2): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 9,086 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 8,889 works ## Cite Cite as: Roy Bar-Haim, Indrajit Bhattacharya, Francesco Dinuzzo, Amrita Saha and Noam Slonim. 2017. Stance Classification of Context-Dependent Claims. Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers. The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{barhaim2017stance, title = {{Stance Classification of Context-Dependent Claims}}, author = {Roy Bar-Haim, Indrajit Bhattacharya, Francesco Dinuzzo, Amrita Saha and Noam Slonim}, booktitle = {Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers}, year = {2017}, url = {https://aclanthology.org/E17-1024/} } ``` - [← previous Speech Recognition Challenge in the Wild: Arabic MGB-3](https://md.research.elamur.ai/free/en/papers/ali-2017-speech-recognition-challenge-wild-arabic/) - [next → Categorical or gradient? An ultrasound investigation of /l/-darkening and vocalization in varieties of English](https://md.research.elamur.ai/free/en/papers/turton-2017-categorical-gradient-ultrasound-investigation-l/) - [↑ back to papers of 2017](https://md.research.elamur.ai/free/en/papers/2017/page/3/) ## References 3 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Ehud Aharoni, Anatoly Polnarov, Tamar Lavee, Daniel Hershcovich, Ran Levy, Ruty Rinott, Dan Gutfreund, and Noam Slonim. 2014. [A benchmark dataset for automatic detection of claims and evidence in the context of controversial topics](https://md.research.elamur.ai/free/en/papers/aharoni-2014-benchmark-dataset-automatic-detection-claims/). In Proceedings of the First Workshop on Argumentation Mining, pages… 2. Overall, the results confirm that our modular approach outperforms the common practice of monolithic classifiers for stance classification, in particular for making high-accuracy stance predictions for a large portion of the claims. Each component was shown to contribute to the overall performance. [find](https://scholar.google.com/scholar?q=Overall%2C+the+results+confirm+that+our+modular+approach+outperforms+the+common+practice+of+monolithic+classifiers+for+stance+classification%2C+in+particular+for+making+high%2Daccuracy+stance+predictions+fo) 3. Clinton Burfoot, Steven Bird, and Timothy Baldwin. 2011. Collective classification of congressional floor-debate transcripts. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, pages 1506–1515, Portland, Oregon, USA, June. Associa… [find](https://scholar.google.com/scholar?q=Clinton+Burfoot%2C+Steven+Bird%2C+and+Timothy+Baldwin%2E+2011%2E+Collective+classification+of+congressional+floor%2Ddebate+transcripts%2E+In+Proceedings+of+the+49th+Annual+Meeting+of+the+Association+for+Computati) *Updated 24 October 2026*