SentiWordNet It (swn_it)

A sentiment lexical resource for Italian, based on SentiWordNet 3.0 and MultiWordNet.

Description

This repository contains a sentiment lexicon for Italian, based on SentiWordNet 3.0 (Baccianella, Esuli, and Sebastiani 2010; Esuli [2019] 2025) and MultiWordNet (Pianta, Bentivogli, and Girardi 2002).

Unlike previous resources—SentiWordNet, which provides sentiment scores without Italian lexical coverage, and MultiWordNet, which offers Italian synsets without sentiment annotation—this dataset bridges the two by mapping Italian lexical entries to sentiment scores in a ready-to-use CSV format.

This integration enables direct use in sentiment analysis and other NLP applications for Italian, filling a gap in existing resources.

The included files, in the data/ folder are:

• swn_it.csv: A dataset of 35,001 Italian synsets with polarity scores, POS, synset, offset, English synset lemmas, and gloss (in English).
• swn_it_tidy.csv: A tidy (one token per row) dataset of 41,725 lemmas, with polarity scores. It is designed for use in R.

It also contains a folder with examples in R, and scripts to use and manipulate the datasets:

• examples-R/:
  • custom_dataset.R: Create a custom tidy dataset from the original one, for treating duplicate entries differently.
  • example.R: Examples of how to use the dataset for sentiment analysis on a sample text.
  • uso.md: Instructions for using the dataset in R (in Italian), referred to in example.R.

Methodology

Sources

• SentiWordNet, based on WordNet 3.0. Repository: https://github.com/aesuli/SentiWordNet; License: CC BY-SA 3.0
• MultiWordNet (Italian), aligned with WordNet 1.6. Website: https://nlplab.fbk.eu/tools-and-resources/lexical-resources-and-corpora/multiwordnet; License: CC BY 3.0

Access and alignment of these resources were performed using NLTK and the library OMW - Open Multilingual Wordnet (Bond et al. 2023).

Steps

The lexicon was generated following these steps:

• Synset Mapping: Each Italian synset in MultiWordNet was aligned with its counterpart in SentiWordNet 3.0 using the synset name (e.g., “a_lot.r.01”).
• Score Assignment: The scores from each SentiWordNet synset were then assigned to the corresponding Italian synset.
• In the Italian lemmas field, the placeholders used in MultiWordNet to mark so-called “lexical gaps” (“GAP!” and “PSEUDOGAP!”) have been retained. These indicate that the synset is not “missing”: it was handled in MultiWordNet but has no lexical equivalent in Italian (Pianta, Bentivogli, and Girardi 2002).
• The synset terms are present in both languages (Italian and English) to facilitate any future comparisons.

Dataset Format

The column naming follows the conventions of the NLTK package for WordNet and SentiWordNet.

Synset Dataset (swn_it.csv)

• Encoding: UTF-8
• Separator: Comma (,)

Columns:

• pos: The part of speech of the synset. Possible values are:
  • n: noun
  • v: verb
  • a: adjective
  • r: adverb
• offset: The unique numerical identifier of a synset within a specific version of WordNet. It is the recommended ID for join and lookup operations within the same version and is used together with the pos.
• synset: The symbolic and human-readable name (lemma.pos.sense_number). This identifier is more stable across versions than the offset, which may be reassigned during WordNet updates (cf. Kafe 2017). This information may be important, considering that MultiWordNet has not been updated beyond version 1.6 of WordNet (cf. Basile and Nissim 2013).
• pos_score: The positivity score of the synset, ranging from 0.0 to 1.0.
• neg_score: The negativity score of the synset, ranging from 0.0 to 1.0.
• obj_score: The objectivity score of the synset, ranging from 0.0 to 1.0. This score is calculated as: obj_score = 1.0 - (pos_score + neg_score).
• lemmi_it: Italian lemmas (terms) belonging to the synset, separated by commas (with original capitalization).
• lemmi_en: English lemmas (terms) belonging to the synset, separated by commas (with original capitalization).
• gloss: The definition of the synset, providing semantic context for the associated lemmas.

“Tidy” Lemma Dataset (swn_it_tidy.csv)

Each row represents a single term (lemma) with its corresponding scores. Capitalization and duplicate entries have been removed by calculating the average scores for each repeated lemma. Consequently, the pos tag no longer applies.

• Encoding: UTF-8
• Separator: Comma (,)

Columns:

• lemma: The term (word) in Italian, in lowercase.
• pos_score: The positivity score of the synset, ranging from 0.0 to 1.0.
• neg_score: The negativity score of the synset, ranging from 0.0 to 1.0.
• obj_score: The objectivity score of the synset, ranging from 0.0 to 1.0.This score is calculated as: obj_score = 1.0 - (pos_score + neg_score).

However, it is possible to reconstruct the tidy dataset from the original dataset, treating duplicate entries differently.

See the script custom_dataset.R in the examples-R folder for details.

How to Cite

If you use sentiwordnet_it 1.0 in your research, please cite it as follows:

Agnese Vardanega. (2025). sentiwordnet_it 1.0 (v1.0.1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.17248244

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY SA 4.0) License.

References

Baccianella, Stefano, Andrea Esuli, and Fabrizio Sebastiani. 2010. “Sentiwordnet 3.0: An Enhanced Lexical Resource for Sentiment Analysis and Opinion Mining.” In Lrec, 10:2200–2204. 2010. Valletta. http://lrec-conf.org/proceedings/lrec2010/pdf/769_Paper.pdf.

Basile, Valerio, and Malvina Nissim. 2013. “Sentiment Analysis on Italian Tweets.” In Proceedings of the 4th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, 100–107. https://aclanthology.org/W13-1614/.

Bond, Francis, Michael Wayne Goodman, Ewa Rudnicka, Luis Morgado da Costa, Alexandre Rademaker, and John P. McCrae. 2023. “Documenting the Open Multilingual Wordnet.” In Proceedings of the 12th Global Wordnet Conference, edited by German Rigau, Francis Bond, and Alexandre Rademaker, 150–57. University of the Basque Country, Donostia - San Sebastian, Basque Country: Global Wordnet Association. https://aclanthology.org/2023.gwc-1.18/.

Denecke, Kerstin. 2008. “Using Sentiwordnet for Multilingual Sentiment Analysis.” In 2008 IEEE 24th International Conference on Data Engineering Workshop, 507–12. IEEE. https://ieeexplore.ieee.org/abstract/document/4498370/.

Esuli, Andrea. (2019) 2025. “Aesuli/SentiWordNet.” https://github.com/aesuli/SentiWordNet.

Kafe, Eric. 2017. “How Stable Are WordNet Synsets?” In LDK Workshops, 113–24. https://ceur-ws.org/Vol-1899/CfWNs_2017_proc1-paper_1.pdf.

Pianta, Emanuele, Luisa Bentivogli, and Christian Girardi. 2002. “MultiWordNet: Developing an Aligned Multilingual Database.” In First International Conference on Global WordNet, 293–302. https://cris.fbk.eu/handle/11582/499.