Portuguese Word Embeddings: Evaluating on Word Analogies and Natural Language Tasks

Hartmann, Nathan Siegle; Fonseca, Erick Rocha; Shulby, Christopher; Treviso, Marcos; Rodrigues, Jéssica Ferreira; Aluísio, Sandra Maria

doi:10.48550/arxiv.1708.06025

Cited by 19 publications

(32 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Such datasets are usually composed of all the possible combinations of pairs such as Paris : France, Berlin : Germany or Beijing : China. In our evaluation, we use the dataset of Svoboda and Brychcin (2016) for Czech, that of Köper et al (2015) for German, that of Cardellino (2016) for Spanish, that of Venekoski and Vankka (2017) for Finnish, that of Berardi et al (2015) for Italian, the European variant of the dataset proposed by Hartmann et al (2017) for Portuguese and that of Chen et al (2015) for Chinese.…”

Section: Evaluation Datasetsmentioning

confidence: 99%

Learning Word Vectors for 157 Languages

Grave¹,

Bojanowski²,

Gupta³

et al. 2018

Preprint

View full text Add to dashboard Cite

Distributed word representations, or word vectors, have recently been applied to many tasks in natural language processing, leading to state-of-the-art performance. A key ingredient to the successful application of these representations is to train them on very large corpora, and use these pre-trained models in downstream tasks. In this paper, we describe how we trained such high quality word representations for 157 languages. We used two sources of data to train these models: the free online encyclopedia Wikipedia and data from the common crawl project. We also introduce three new word analogy datasets to evaluate these word vectors, for French, Hindi and Polish. Finally, we evaluate our pre-trained word vectors on 10 languages for which evaluation datasets exists, showing very strong performance compared to previous models.

show abstract

Section: Evaluation Datasetsmentioning

confidence: 99%

Learning Word Vectors for 157 Languages

Grave¹,

Bojanowski²,

Gupta³

et al. 2018

Preprint

View full text Add to dashboard Cite

show abstract

“…To do so, we relied upon pretrained language models. Specifically, since the tweets in our dataset were written in Brazilian Portuguese, we downloaded several pretrained Brazilian Portuguese corpora [6], including those trained using two different variations of the Word2Vec algorithm: a) Continuous Bag-of-Words (CBOW), and b) Skip-Gram with Negative Sampling (SGNS), and GloVe [7]. All corpora used 300 dimensions.…”

Section: Methodsmentioning

confidence: 99%

“…The specific words used to calculate purity were derived from the Brazilian Portuguese Moral Foundations Dictionary for Fake News classification [11]. We retained all words from this dictionary if they were present in our pretrained Brazilian Portuguese corpora [6]. In each case, words were selected based upon their presence in these word lists, with words only excluded if they did not appear in our pretrained corpus or the corresponding tweets.…”

Section: Methodsmentioning

confidence: 99%

Applying Word Embeddings to Measure Valence in Information Operations Targeting Journalists in Brazil

Broniatowski

2022

Preprint

View full text Add to dashboard Cite

Among the goals of information operations are to change the overall information environment vis-à-vis specific actors. For example, "trolling campaigns" seek to undermine the credibility of specific public figures, leading others to distrust them and intimidating these figures into silence. To accomplish these aims, information operations frequently make use of "trolls" -malicious online actors who target verbal abuse at these figures. In Brazil, in particular, allies of Brazil's current president have been accused of operating a "hate cabinet" -a trolling operation that targets journalists who have alleged corruption by this politician and other members of his regime. Leading approaches to detecting harmful speech, such as Google's Perspective API, seek to identify specific messages with harmful content. While this approach is helpful in identifying content to downrank, flag, or remove, it is known to be brittle, and may miss attempts to introduce more subtle biases into the discourse. Here, we aim to develop a measure that might be used to assess how targeted information operations seek to change the overall valence, or appraisal, of specific actors. Preliminary results suggest known campaigns target female journalists more so than male journalists, and that these campaigns may leave detectable traces in overall Twitter discourse.

show abstract

“…A large Portuguese corpus was gathered and described in Hartmann et al (2017) , including both Brazilian and European variants. It was used for training and evaluating different word embedding models (FastText, Glove, Wang2Vec and Word2Vec).…”

Section: Related Workmentioning

confidence: 99%

“…However, the great majority of such models has been developed for English corpora. It was only in recent years that the research community has also been focusing on other languages with rich morphology and different syntaxes ( Hartmann et al, 2017 ; Rodrigues et al, 2016 ; Sun et al, 2016 ; Svoboda & Beliga, 2017 ; Svoboda & Brychcin, 2016 ; Turian, Ratinov & Bengio, 2010 ). Moreover, large datasets are required to enable achieving good performance.…”

Section: Introductionmentioning

confidence: 99%

Improving word embeddings in Portuguese: increasing accuracy while reducing the size of the corpus

Pinto

Viana

Teixeira

et al. 2022

PeerJ Computer Science

View full text Add to dashboard Cite

The subjectiveness of multimedia content description has a strong negative impact on tag-based information retrieval. In our work, we propose enhancing available descriptions by adding semantically related tags. To cope with this objective, we use a word embedding technique based on the Word2Vec neural network parameterized and trained using a new dataset built from online newspapers. A large number of news stories was scraped and pre-processed to build a new dataset. Our target language is Portuguese, one of the most spoken languages worldwide. The results achieved significantly outperform similar existing solutions developed in the scope of different languages, including Portuguese. Contributions include also an online application and API available for external use. Although the presented work has been designed to enhance multimedia content annotation, it can be used in several other application areas.

show abstract

Portuguese Word Embeddings: Evaluating on Word Analogies and Natural Language Tasks

Cited by 19 publications

References 9 publications

Learning Word Vectors for 157 Languages

Learning Word Vectors for 157 Languages

Applying Word Embeddings to Measure Valence in Information Operations Targeting Journalists in Brazil

Improving word embeddings in Portuguese: increasing accuracy while reducing the size of the corpus

Contact Info

Product

Resources

About