---
title: "SentiWordNet 3.0 — Baccianella 2010"
description: "The paper describes SentiWordNet 3.0, a lexicon that gives every WordNet 3.0 synset three scores, Pos, Neg and Obj, which add up to 1."
authors: ["Stefano Baccianella", "Andrea Esuli", "Fabrizio Sebastiani"]
year: 2010
venue: "Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC'10)"
source: "https://aclanthology.org/L10-1531/"
licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)"
licence_url: "https://creativecommons.org/licenses/by-nc/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/papers/baccianella-2010-sentiwordnet-3-0-enhanced-lexical/"
---
[Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/baccianella-2010-sentiwordnet-3-0-enhanced-lexical/)

[home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [2010](https://md.research.elamur.ai/free/papers/2010/) / SentiWordNet 3.0: An Enhanced Lexical Resource for Sentiment Analysis and Opinion Mining

# SentiWordNet 3.0: An Enhanced Lexical Resource for Sentiment Analysis and Opinion Mining

**Authors:** Stefano Baccianella, [Andrea Esuli](https://md.research.elamur.ai/free/authors/andrea-esuli/), [Fabrizio Sebastiani](https://md.research.elamur.ai/free/authors/fabrizio-sebastiani/) · **Year:** 2010 · **Venue:** Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC'10) · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/L10-1531/) · [PDF](https://aclanthology.org/L10-1531.pdf) · [OpenAlex](https://openalex.org/W193524605)

**Citations:** 2666 citations (openalex-snapshot · 2026-10-14) · [402](#cited-by) in our library, of the 140673 papers whose bibliographies we read

## Summary
The paper describes SentiWordNet 3.0, a lexicon that gives every WordNet 3.0 synset three scores, Pos, Neg and Obj, which add up to 1. Because the scores belong to senses rather than words, different meanings of the same word can carry different sentiment. The resource updates SentiWordNet 1.0, which was built on WordNet 2.0 and is licensed to more than 300 research groups. Annotation happens in two stages. First, a weakly supervised step expands small seed sets and uses them to train a committee of gloss classifiers. Glosses are represented as sequences of synsets, taken from the manually disambiguated Princeton WordNet Gloss Corpus. Second, a random walk over a graph linking each gloss word to the synset it defines refines the scores. To evaluate the resource, the authors mapped a hand-annotated sample onto WordNet 3.0 and compared rankings with a tie-aware Kendall distance. Version 3.0 shows about 20% relative error reduction over 1.0, and the random-walk stage alone accounts for 17–19% of the improvement. This matters because the lexicon is widely used as a ready-made source of sentiment priors in opinion mining.

## Key points
- **Seed expansion.** Seeds are the synsets of 7 typically positive and 7 typically negative terms, following Turney & Littman. They are expanded through WordNet relations that either keep polarity (e.g., also-see) or flip it (direct antonymy), up to a radius k.
- **Classifier committee.** The committee has 8 ternary gloss classifiers: radius k ∈ {0, 2, 4, 6} combined with Rocchio or SVMs. Each final score is the average across all 8 members.
- **Score rescaling.** The raw random-walk values are tiny; the top positive synset gets about 8.3·10⁻⁶. The authors therefore fit a power law a·x^b to the score distribution from the semi-supervised step and apply it to the random-walk ranking. Obj fills the remainder so the three scores sum to 1.
- **Evaluation data.** The test set is Micro-WN(Op), 1,105 manually annotated synsets, mapped automatically to WordNet 3.0 and released as Micro-WN(Op)-3.0. The metric is p-normalized Kendall τ with p = 1/2, where lower is better.
- **Results.**

  | Comparison | Positivity τp | Negativity τp |
  |---|---|---|
  | SentiWordNet 1.0 → 3.0 | .349 → .281 (−19.48%) | .296 → .231 (−21.96%) |
  | 3.0-semi → 3.0 | .339 → .281 (−17.11%) | .286 → .231 (−19.23%) |

## Interesting ideas
- **Polarity-aware graph expansion.** Relations are split into those that keep sentiment and those that invert it, so a handful of seed words can grow into a sizeable training set. The same trick could bootstrap lexicons for languages with few resources, such as a Ukrainian WordNet-like graph.
- **Definitions as a propagation graph.** Sentiment spreads from the words used in a definition to the term being defined. Any dictionary with sense-linked definitions could support this kind of label propagation.
- **Borrowing a distribution from a trusted step.** The random walk orders synsets well but produces unusable magnitudes. The authors keep the ranking and map it onto a power-law curve fitted to the earlier, better-calibrated scores. This is a general way to turn a good ranker into usable graded scores.
- **Committee over hyperparameter settings.** Averaging classifiers trained with different seed radii and different learners smooths out the arbitrary choice of a single setting. This is a cheap way to make weakly supervised labels more robust.
- **Ranking-based evaluation with tie handling.** Graded lexicons are judged by how well their orderings agree with a gold standard. Predicted ties are charged half a penalty, the cost of a random guess. This is useful wherever exact score values are less meaningful than relative order.

*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/)

## Cited by

Cited by

402 papers of our library cite it (of the 140673 whose bibliographies we read); 4 of them have a page here, 398 not yet.

- [CILI: the Collaborative Interlingual Index](https://md.research.elamur.ai/free/papers/bond-2016-cili-collaborative-interlingual-index/)
- [“When Numbers Matter!”: Detecting Sarcasm in Numerical Portions of Text](https://md.research.elamur.ai/free/papers/dubey-2019-when-numbers-matter-detecting-sarcasm/)
- [Semantic Change Characterization with LLMs using Rhetorics](https://md.research.elamur.ai/free/papers/sa-2026-semantic-change-characterization-llms-rhetorics/)
- [Lexicon-Based Methods for Sentiment Analysis](https://md.research.elamur.ai/free/papers/taboada-2011-lexicon-based-methods-sentiment-analysis/)

## Cite

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

```bibtex
@inproceedings{baccianella2010sentiwordnet,
  title = {{SentiWordNet 3.0: An Enhanced Lexical Resource for Sentiment Analysis and Opinion Mining}},
  author = {Baccianella, Stefano and Esuli, Andrea and Sebastiani, Fabrizio},
  booktitle = {Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC'10)},
  year = {2010},
  doi = {10.63317/2jktnrykui9s},
  url = {https://doi.org/10.63317/2jktnrykui9s}
}
```

- [← previous Semi-Supervised Recognition of Sarcasm in Twitter and Amazon](https://md.research.elamur.ai/free/papers/davidov-2010-semi-supervised-recognition-sarcasm-twitter/)
- [next → Simultaneous Tokenization and Part-Of-Speech Tagging for Arabic without a Morphological Analyzer](https://md.research.elamur.ai/free/papers/aclbulk-2010-kulick-simultaneous-tokenization-part-speech-tagging/)
- [↑ back to papers of 2010](https://md.research.elamur.ai/free/papers/2010/)

## References

9 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. S. Cerini, V. Compagnoni, A. Demontis, M. Formentelli, and G. Gandini. 2007. Micro-WNOp: A gold standard for the evaluation of automatically compiled lexical resources for opinion mining. In Andrea Sansò, editor, Language resources and linguistic theory: Typology, second language acquisition, Engli… [find](https://scholar.google.com/scholar?q=S%2E+Cerini%2C+V%2E+Compagnoni%2C+A%2E+Demontis%2C+M%2E+Formentelli%2C+and+G%2E+Gandini%2E+2007%2E+Micro%2DWNOp%3A+A+gold+standard+for+the+evaluation+of+automatically+compiled+lexical+resources+for+opinion+mining%2E+In+Andrea+Sa)
2. Andrea Esuli and Fabrizio Sebastiani. 2006. S ENTI - W ORD N ET : A publicly available lexical resource for opinion mining. In Proceedings of the 5th Conference on Language Resources and Evaluation (LREC’06), pages 417–422, Genova, IT. [find](https://scholar.google.com/scholar?q=Andrea+Esuli+and+Fabrizio+Sebastiani%2E+2006%2E+S+ENTI+%2D+W+ORD+N+ET+%3A+A+publicly+available+lexical+resource+for+opinion+mining%2E+In+Proceedings+of+the+5th+Conference+on+Language+Resources+and+Evaluation+%28L)
3. Andrea Esuli and Fabrizio Sebastiani. 2007a. Random-walk models of term semantics: An application to opinion-related properties. In Proceedings of the 3rd Language Technology Conference (LTC’07), pages 221– 225, Poznań, PL. [find](https://scholar.google.com/scholar?q=Andrea+Esuli+and+Fabrizio+Sebastiani%2E+2007a%2E+Random%2Dwalk+models+of+term+semantics%3A+An+application+to+opinion%2Drelated+properties%2E+In+Proceedings+of+the+3rd+Language+Technology+Conference+%28LTC%E2%80%9907%29%2C+page)
4. Andrea Esuli and Fabrizio Sebastiani. 2007b. S ENTI - W ORD N ET : A high-coverage lexical resource for opinion mining. Technical Report 2007-TR-02, Istituto di Scienza e Tecnologie dell’Informazione, Consiglio Nazionale delle Ricerche, Pisa, IT. [find](https://scholar.google.com/scholar?q=Andrea+Esuli+and+Fabrizio+Sebastiani%2E+2007b%2E+S+ENTI+%2D+W+ORD+N+ET+%3A+A+high%2Dcoverage+lexical+resource+for+opinion+mining%2E+Technical+Report+2007%2DTR%2D02%2C+Istituto+di+Scienza+e+Tecnologie+dell%E2%80%99Informazione%2C)
5. Andrea Esuli. 2008. Automatic Generation of Lexical Resources for Opinion Mining: Models, Algorithms, and Applications. Ph.D. thesis, Scuola di Dottorato in Ingegneria ”Leonardo da Vinci”, University of Pisa, Pisa, IT. [find](https://scholar.google.com/scholar?q=Andrea+Esuli%2E+2008%2E+Automatic+Generation+of+Lexical+Resources+for+Opinion+Mining%3A+Models%2C+Algorithms%2C+and+Applications%2E+Ph%2ED%2E+thesis%2C+Scuola+di+Dottorato+in+Ingegneria+%E2%80%9DLeonardo+da+Vinci%E2%80%9D%2C+University+)
6. Ronald Fagin, Ravi Kumar, Mohammad Mahdiany, D. Sivakumar, and Erik Veez. 2004. Comparing and aggregating rankings with ties. In Proceedings of ACM International Conference on Principles of Database Systems (PODS’04), pages 47–58, Paris, FR. [find](https://scholar.google.com/scholar?q=Ronald+Fagin%2C+Ravi+Kumar%2C+Mohammad+Mahdiany%2C+D%2E+Sivakumar%2C+and+Erik+Veez%2E+2004%2E+Comparing+and+aggregating+rankings+with+ties%2E+In+Proceedings+of+ACM+International+Conference+on+Principles+of+Database+S)
7. Sanda M. Harabagiu, George A. Miller, and Dan I. Moldovan. 1999. WordNet 2: A morphologically and semantically enhanced resource. In Proceedings of the ACL Workshop on Standardizing Lexical Resources (SIGLEX’99), pages 1–8, College Park, US. [find](https://scholar.google.com/scholar?q=Sanda+M%2E+Harabagiu%2C+George+A%2E+Miller%2C+and+Dan+I%2E+Moldovan%2E+1999%2E+WordNet+2%3A+A+morphologically+and+semantically+enhanced+resource%2E+In+Proceedings+of+the+ACL+Workshop+on+Standardizing+Lexical+Resources+)
8. Bo Pang and Lillian Lee. 2008. Opinion mining and sentiment analysis. Foundations and Trends in Information Retrieval, 2(1/2):1–135. [find](https://scholar.google.com/scholar?q=Bo+Pang+and+Lillian+Lee%2E+2008%2E+Opinion+mining+and+sentiment+analysis%2E+Foundations+and+Trends+in+Information+Retrieval%2C+2%281%2F2%29%3A1%E2%80%93135%2E)
9. Peter D. Turney and Michael L. Littman. 2003. Measuring praise and criticism: Inference of semantic orientation from association. ACM Transactions on Information Systems, 21(4):315–346. [find](https://scholar.google.com/scholar?q=Peter+D%2E+Turney+and+Michael+L%2E+Littman%2E+2003%2E+Measuring+praise+and+criticism%3A+Inference+of+semantic+orientation+from+association%2E+ACM+Transactions+on+Information+Systems%2C+21%284%29%3A315%E2%80%93346%2E)