---
title: "ISWARA at WNUT-2020 Task 2 — paper by Putri et al. (2020)"
description: "The paper describes team ISWARA's entry to WNUT-2020 Task 2. 0.763 precision. 8 references."
authors: ["Wava Carissa Putri", "Rani Aulia Hidayat", "Isnaini Nurul Khasanah", "Rahmad Mahendra"]
year: 2020
venue: "Proceedings of the Sixth Workshop on Noisy User-generated Text (W-NUT 2020)"
source: "https://aclanthology.org/2020.wnut-1.74/"
doi: "10.18653/v1/2020.wnut-1.74"
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/putri-2020-iswara-wnut-2020-task-2/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/putri-2020-iswara-wnut-2020-task-2/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2020](https://md.research.elamur.ai/free/en/papers/2020/page/10/) / ISWARA at WNUT-2020 Task 2: Identification of Informative COVID-19 English Tweets using BERT and FastText Embeddings

# ISWARA at WNUT-2020 Task 2: Identification of Informative COVID-19 English Tweets using BERT and FastText Embeddings

**Authors:** Wava Carissa Putri, Rani Aulia Hidayat, Isnaini Nurul Khasanah and [Rahmad Mahendra](https://md.research.elamur.ai/free/en/authors/rahmad-mahendra/) · **Year:** 2020 · **Venue:** [Proceedings of the Sixth Workshop on Noisy User-generated Text (W-NUT 2020)](https://md.research.elamur.ai/free/en/venues/wnut/) ([Workshop on Noisy User-generated Text 2020](https://md.research.elamur.ai/free/en/venues/wnut/2020/)) · **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/2020.wnut-1.74/) · [PDF](https://aclanthology.org/2020.wnut-1.74.pdf) · [DOI 10.18653/v1/2020.wnut-1.74](https://doi.org/10.18653/v1/2020.wnut-1.74) · [OpenAlex](https://openalex.org/W3098683905)

**Citations:** 1 citations (OpenAlex, 2026-10-14)

## Summary
The paper describes team ISWARA's entry to WNUT-2020 Task 2. The task is binary classification of English COVID-19 tweets as informative or not. A tweet counts as informative when it reports recovered, suspected, confirmed or death cases. The organizers supplied 7,000 training tweets and 1,000 validation tweets. After cleaning, the authors tested two pipelines:

- **DistilBERT pipeline:** sentence representations from DistilBERT went into Logistic Regression or an RBF-kernel SVM. Some runs also added a small set of hand-picked keyword-occurrence features and a flag for whether the tweet contains a number.
- **fastText pipeline:** the supervised fastText classifier (multinomial logistic regression) was trained directly on the cleaned text.

On validation data, fastText and DistilBERT+keywords+SVM tied at F1 0.81. On the official test set, the BERT-based system came out ahead, with F1 0.763 against 0.757. The paper shows that cheap, definition-driven lexical cues can lift a generic embedding classifier. It also shows that a validation tie can resolve differently on unseen data.

## Key points
- **Data:** the training set has 3,697 uninformative and 3,303 informative tweets. The validation set has 528 uninformative and 472 informative.
- **Preprocessing:** URLs, non-ASCII characters, "RT", "@", "&", "\<", ">" and extra spaces were removed, and all text was lowercased.
- **Keyword features:** 'corona', 'covid', 'recover', 'suspect', 'confirm', 'death' and 'case', plus a binary flag for the presence of a numeral.
- **Classifier settings:**
  - LR: C ≈ 5.26, up to 10,000 iterations.
  - SVM: RBF kernel, C=10, gamma=0.01.
  - fastText: lr 0.075, word n-grams 2, 150 epochs, minCount 6, 200,000 buckets, hierarchical softmax loss. Manually chosen values beat automatic tuning.
- **Test results for the best system (BERT & WO + SVM):** F1 0.763, precision 0.807, recall 0.724, accuracy 0.788. fastText+MLR scored F1 0.757. Keyword features improved BERT with both LR and SVM.

## Interesting ideas
- Turning the annotation guideline itself into a handful of indicator features (the case-type words named in the definition) gives a classifier a direct, inspectable link to what the label means.
- Plotting how often each candidate cue occurs per class showed that only some keywords actually separate the classes ('confirm', 'death', 'case'). Others, like 'covid' or numbers, were near-uniform and contributed little on their own.
- A simple "contains a number" flag captures the fact that informative reports usually state counts.
- Two systems tied on validation but ranked differently on the test set, which the authors attribute to subword handling of unseen words.
- The team picked a distilled transformer because it was much smaller at similar quality.

## For Mova

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

- **I1** (frames): lemma-level trigger lists taken from frame or speech-act definitions could act as transparent priors for detecting report-type frames (e.g., "confirm", "die"), with each firing logged as an explanation.
- **I2** (learn): a per-feature class-distribution audit can serve as a cheap check that a repair rule or cue really discriminates rather than simply correlating with the topic.
- **I3** (world): numerals attached to case-like nouns could be promoted to quantity states of entities, so that count-reporting sentences populate measures in the text's world.
- **I4** (learn): model selection should not rest on a single dev split, and an out-of-vocabulary-heavy held-out slice should be part of evaluation.
- **I5** (tree): distilled encoders are a reasonable first choice for gparse when GPU budget matters, provided LAS is checked against the full-size model.

*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 (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

🧭 Topics (1): [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,263 works

## Cite

Cite as: Wava Carissa Putri, Rani Aulia Hidayat, Isnaini Nurul Khasanah and Rahmad Mahendra. 2020. ISWARA at WNUT-2020 Task 2: Identification of Informative COVID-19 English Tweets using BERT and FastText Embeddings. Proceedings of the Sixth Workshop on Noisy User-generated Text (W-NUT 2020).

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

```bibtex
@inproceedings{putri2020iswara,
  title = {{ISWARA at WNUT-2020 Task 2: Identification of Informative COVID-19 English Tweets using BERT and FastText Embeddings}},
  author = {Wava Carissa Putri, Rani Aulia Hidayat, Isnaini Nurul Khasanah and Rahmad Mahendra},
  booktitle = {Proceedings of the Sixth Workshop on Noisy User-generated Text (W-NUT 2020)},
  year = {2020},
  doi = {10.18653/v1/2020.wnut-1.74},
  url = {https://doi.org/10.18653/v1/2020.wnut-1.74}
}
```

- [← previous IRLab@IITBHU at WNUT-2020 Task 2: Identification of informative COVID-19 English Tweets using BERT](https://md.research.elamur.ai/free/en/papers/chanda-2020-irlab-iitbhu-wnut-2020-task/)
- [next → Measuring Malaysian Students Achievement Level in Arabic Morphology: Improving and Developing Curriculum](https://md.research.elamur.ai/free/en/papers/sidiq-2020-measuring-malaysian-students-achievement-level/)
- [↑ back to papers of 2020](https://md.research.elamur.ai/free/en/papers/2020/page/10/)

## References

8 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Ashktorab, Z., Brown, C., Nandi, M., & Culotta, A. 2014 . Tweedr: Mining twitter to inform disaster response. In Information Systems for Crisis Response and Management (ISCRAM), pages 269-272. [find](https://scholar.google.com/scholar?q=Ashktorab%2C+Z%2E%2C+Brown%2C+C%2E%2C+Nandi%2C+M%2E%2C+%26+Culotta%2C+A%2E+2014+%2E+Tweedr%3A+Mining+twitter+to+inform+disaster+response%2E+In+Information+Systems+for+Crisis+Response+and+Management+%28ISCRAM%29%2C+pages+269%2D272%2E)
2. Kevin Roitero, Cristian Bozzato, Vincenzo Della Mea, Steffano Mizzaro, and Giuseppe Serra. 2020. Twitter goes to the Doctor: Detecting Medical Tweets using Machine Learning and BERT. In Proceedings of the International Workshop on Semantic Indexing and Information Retrieval for Health from heterogeneous content types and languages (SIIRH) 2020. [find](https://scholar.google.com/scholar?q=Kevin+Roitero%2C+Cristian+Bozzato%2C+Vincenzo+Della+Mea%2C+Steffano+Mizzaro%2C+and+Giuseppe+Serra%2E+2020%2E+Twitter+goes+to+the+Doctor%3A+Detecting+Medical+Tweets+using+Machine+Learning+and+BERT%2E+In+Proceedings+of)
3. Vieweg, S. 2010. Microblogged contributions to the emergency arena: Discovery, interpretation and implications. Computer Supported Collaborative Work, pages 515-516. [find](https://scholar.google.com/scholar?q=Vieweg%2C+S%2E+2010%2E+Microblogged+contributions+to+the+emergency+arena%3A+Discovery%2C+interpretation+and+implications%2E+Computer+Supported+Collaborative+Work%2C+pages+515%2D516%2E)
4. Dat Quoc Nguyen, Thanh Vu, Afshin Rahimi, Mai Hoang Dao, Linh The Nguyen, and Long Doan. 2020. [WNUT-2020 Task 2: Identification of Informative COVID-19 English Tweets](https://md.research.elamur.ai/free/en/papers/nguyen-2020-wnut-2020-task-2-identification/). In Proceedings of the 6th Workshop Noisy User-generated Text.
5. Ali Alessa, Miad Faezipour, and Zakhriya Alhassan. 2018. Text classification of flu-related tweets using fasttext with sentiment and keyword features. 2018 IEEE International Conference on Healthcare Informatics (ICHI), pages 366-367. [find](https://scholar.google.com/scholar?q=Ali+Alessa%2C+Miad+Faezipour%2C+and+Zakhriya+Alhassan%2E+2018%2E+Text+classification+of+flu%2Drelated+tweets+using+fasttext+with+sentiment+and+keyword+features%2E+2018+IEEE+International+Conference+on+Healthcare+)
6. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://md.research.elamur.ai/free/en/papers/devlin-2019-bert-pre-training-deep-bidirectional/). arXiv preprint arXiv:1810.04805. [↗](https://arxiv.org/abs/1810.04805)
7. Armand Joulin, Edouard Grave, Piotr Bojanowski, and Tomas Mikolov. 2017. Bag of Tricks for Efficient Text Classification. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics (Volume 2: Short Papers). Association for Computational Linguistics, pages 427-431. [find](https://scholar.google.com/scholar?q=Armand+Joulin%2C+Edouard+Grave%2C+Piotr+Bojanowski%2C+and+Tomas+Mikolov%2E+2017%2E+Bag+of+Tricks+for+Efficient+Text+Classification%2E+In+Proceedings+of+the+15th+Conference+of+the+European+Chapter+of+the+Associati)
8. Martin Müller, Marcel Salathé, and Per E Kummervold. 2020. COVID-Twitter-BERT: A Natural Language Processing Model to Analyse 4 494 [find](https://scholar.google.com/scholar?q=Martin+M%C3%BCller%2C+Marcel+Salath%C3%A9%2C+and+Per+E+Kummervold%2E+2020%2E+COVID%2DTwitter%2DBERT%3A+A+Natural+Language+Processing+Model+to+Analyse+4+494)

*Updated 1 November 2026*