---
title: "THU_NGN at SemEval-2018 Task 1 — Wu 2018"
description: "This paper presents the THU_NGN system for SemEval-2018 Task 3, which targets irony in English tweets after explicit irony hashtags have been removed."
authors: ["Chuhan Wu", "Fangzhao Wu", "Junxin Liu", "Zhigang Yuan", "Sixing Wu", "Yongfeng Huang"]
year: 2018
venue: "Proceedings of the 12th International Workshop on Semantic Evaluation"
source: "https://aclanthology.org/S18-1028/"
doi: "10.18653/v1/s18-1028"
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/papers/wu-2018-thu-ngn-semeval-2018-task/"
---
[Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/wu-2018-thu-ngn-semeval-2018-task/)

[home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [2018](https://md.research.elamur.ai/free/papers/2018/page/3/) / THU\_NGN at SemEval-2018 Task 1: Fine-grained Tweet Sentiment Intensity Analysis with Attention CNN-LSTM

# THU\_NGN at SemEval-2018 Task 1: Fine-grained Tweet Sentiment Intensity Analysis with Attention CNN-LSTM

**Authors:** Chuhan Wu, Fangzhao Wu, Junxin Liu, Zhigang Yuan, Sixing Wu, Yongfeng Huang · **Year:** 2018 · **Venue:** Proceedings of the 12th International Workshop on Semantic Evaluation · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/S18-1028/) · [PDF](https://aclanthology.org/S18-1028.pdf) · [DOI 10.18653/v1/s18-1028](https://doi.org/10.18653/v1/s18-1028) · [OpenAlex](https://openalex.org/W2805737268)

**Citations:** 7 citations (openalex-snapshot · 2026-10-14) · 3 in our library, of the 140673 papers whose bibliographies we read

## Summary
This paper presents the THU_NGN system for SemEval-2018 Task 3, which targets irony in English tweets after explicit irony hashtags have been removed. Subtask A is a binary ironic vs. non-ironic decision. Subtask B separates three irony types (V-irony, O-irony, S-irony) from non-ironic tweets.

The core model, Dense-LSTM, stacks bidirectional LSTM layers so that each layer receives the concatenated outputs of every layer below it. This lets the network use shallow and deep contextual features together. Each token is represented by two concatenated pre-trained tweet embeddings plus one-hot POS tags. The final hidden representation is then combined with lexicon-based sentiment features from AffectiveTweets and an averaged sentence embedding.

The network is trained on three related objectives at once:
- recovering the removed irony hashtag,
- the binary irony decision,
- irony-type classification, using a class-weighted loss to handle rare types.

Averaging ten models trained with random dropout gave the final submission. It placed 2nd of 43 teams in subtask A and 3rd of 29 in subtask B. This shows that dense layer connections and auxiliary tasks help detect subtle, hashtag-free irony.

## Key points
- Data: 3,834 training tweets (1,911 ironic, 1,923 non-ironic) and 784 test tweets. The ironic class is unevenly split across types, with V-irony being the most common.
- Setup: 4 Bi-LSTM layers with 200-dim hidden states, trained with RMSProp at batch size 64. Word vectors combine the 400-dim Godin et al. embeddings (trained on 400M tweets) with the 300-dim Barbieri et al. embeddings (trained on 20M tweets). POS tags come from Ark-Tweet-NLP.
- Results: Dense-LSTM+ens reached an F-score of 70.54 in subtask A and a macro-F of 49.47 in subtask B. The SVM bag-of-words baseline scored 58.47 and 32.69.
- Multi-task ablation (without ensembling): using all three objectives raised subtask A F from 65.10 with the binary task alone to 67.36. Subtask B macro-F rose from 44.65 to 48.28.
- Feature ablation: POS tags gave the largest gain among the extra features, and each feature helped. Concatenating both embedding sets beat using either one alone.

*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](https://md.research.elamur.ai/free/shelves/linguistics/) · [Natural language processing](https://md.research.elamur.ai/free/shelves/nlp/)

Topics: [Part-of-speech tagging](https://md.research.elamur.ai/free/topics/part-of-speech-tagging/)

## Cite

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

```bibtex
@inproceedings{wu2018thu,
  title = {{THU\_NGN at SemEval-2018 Task 1: Fine-grained Tweet Sentiment Intensity Analysis with Attention CNN-LSTM}},
  author = {Wu, Chuhan and Wu, Fangzhao and Liu, Junxin and Yuan, Zhigang and Wu, Sixing and Huang, Yongfeng},
  booktitle = {Proceedings of the 12th International Workshop on Semantic Evaluation},
  year = {2018},
  doi = {10.18653/v1/s18-1028},
  url = {https://doi.org/10.18653/v1/s18-1028}
}
```

- [← previous They Exist! Introducing Plural Mentions to Coreference Resolution and Entity Linking](https://md.research.elamur.ai/free/papers/zhou-2018-they-exist-introducing-plural-mentions/)
- [next → Towards a complex analysis of Wayihî + T constructions in Biblical Hebrew](https://md.research.elamur.ai/free/papers/hbo-2018-westbury-towards-complex-analysis-wayihi-t/)
- [↑ back to papers of 2018](https://md.research.elamur.ai/free/papers/2018/page/3/)

## References

7 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Francesco Barbieri, German Kruszewski, Francesco Ronzano, and Horacio Saggion. 2016. How cosmopolitan are emojis?: Exploring emojis usage and Raúl Santos-Rodrı́guez, Darı́o Garcı́a-Garcı́a, and Jesús Cid-Sueiro. 2009. Cost-sensitive classification based on bregman divergences for medical diagnosis… [find](https://scholar.google.com/scholar?q=Francesco+Barbieri%2C+German+Kruszewski%2C+Francesco+Ronzano%2C+and+Horacio+Saggion%2E+2016%2E+How+cosmopolitan+are+emojis%3F%3A+Exploring+emojis+usage+and+Rau%CC%81l+Santos%2DRodr%C4%B1%CC%81guez%2C+Dar%C4%B1%CC%81o+Garc%C4%B1%CC%81a%2DGarc%C4%B1%CC%81a%2C+and+Jesu%CC%81)
2. Mike Thelwall, Kevan Buckley, and Georgios Pal-toglou. 2012. Sentiment strength detection for the social web. Journal of the Association for Information Science and Technology, 63(1):163–173. [find](https://scholar.google.com/scholar?q=Mike+Thelwall%2C+Kevan+Buckley%2C+and+Georgios+Pal%2Dtoglou%2E+2012%2E+Sentiment+strength+detection+for+the+social+web%2E+Journal+of+the+Association+for+Information+Science+and+Technology%2C+63%281%29%3A163%E2%80%93173%2E)
3. Cynthia Van Hee, Els Lefever, and Véronique Hoste. 2016a. Exploring the realization of irony in twitter data. In LREC. [find](https://scholar.google.com/scholar?q=Cynthia+Van+Hee%2C+Els+Lefever%2C+and+Ve%CC%81ronique+Hoste%2E+2016a%2E+Exploring+the+realization+of+irony+in+twitter+data%2E+In+LREC%2E) [our card](https://md.research.elamur.ai/free/papers/hee-2016-exploring-realization-irony-twitter-data/)
4. Cynthia Van Hee, Els Lefever, and Véronique Hoste. 2016b. Monday mornings are my fave:)\# not exploring the automatic recognition of irony in english tweets. In Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2730–2739. [find](https://scholar.google.com/scholar?q=Cynthia+Van+Hee%2C+Els+Lefever%2C+and+Ve%CC%81ronique+Hoste%2E+2016b%2E+Monday+mornings+are+my+fave%3A%29%5C%23+not+exploring+the+automatic+recognition+of+irony+in+english+tweets%2E+In+Proceedings+of+COLING+2016%2C+the+26th+I) [our card](https://md.research.elamur.ai/free/papers/hee-2016-monday-mornings-are-my-fave/)
5. Cynthia Van Hee, Els Lefever, and Véronique Hoste. 2018. SemEval-2018 Task 3: Irony Detection in English Tweets. In Proceedings of the 12th International Workshop on Semantic Evaluation, SemEval-2018, New Orleans, LA, USA. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Cynthia+Van+Hee%2C+Els+Lefever%2C+and+Ve%CC%81ronique+Hoste%2E+2018%2E+SemEval%2D2018+Task+3%3A+Irony+Detection+in+English+Tweets%2E+In+Proceedings+of+the+12th+International+Workshop+on+Semantic+Evaluation%2C+SemEval%2D2018) [our card](https://md.research.elamur.ai/free/papers/hee-2018-semeval-2018-task-3-irony/)
6. Chuhan Wu, Fangzhao Wu, Yongfeng Huang, Sixing Wu, and Zhigang Yuan. 2017. Thu ngn at ijcnlp-2017 task 2: Dimensional sentiment analysis for chinese phrases with deep lstm. Proceedings of the IJCNLP 2017, Shared Tasks, pages 47–52. [find](https://scholar.google.com/scholar?q=Chuhan+Wu%2C+Fangzhao+Wu%2C+Yongfeng+Huang%2C+Sixing+Wu%2C+and+Zhigang+Yuan%2E+2017%2E+Thu+ngn+at+ijcnlp%2D2017+task+2%3A+Dimensional+sentiment+analysis+for+chinese+phrases+with+deep+lstm%2E+Proceedings+of+the+IJCNLP+2)
7. Rui Xia, Chengqing Zong, and Shoushan Li. 2011. Ensemble of feature sets and classification algorithms for sentiment classification. Information Sciences, 181(6):1138–1152. [find](https://scholar.google.com/scholar?q=Rui+Xia%2C+Chengqing+Zong%2C+and+Shoushan+Li%2E+2011%2E+Ensemble+of+feature+sets+and+classification+algorithms+for+sentiment+classification%2E+Information+Sciences%2C+181%286%29%3A1138%E2%80%931152%2E)