Contextual-Lexicon Approach for Abusive Language Detection

Vargas, Francielle Alves; Góes, Fabiana Rodrigues de; Carvalho, Isabelle; Benevenuto, Fabrí­cio; Pardo, Thiago Alexandre Salgueiro

doi:10.26615/978-954-452-072-4_161

Cited by 10 publications

(18 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Examples of works that applied sentiment analysis for hate speech detection were also proposed in (Njagi et al 2015;). ( 4) Lexical resources: In this approach, a controlled vocabulary of hateful and offensive words and expressions are used as features (Xiang et al 2012;Vargas et al 2021;Burnap and Williams 2016). ( 5) Linguistic features: Linguistic information is surely relevant for text classification and have been explored for hate speech detection such as part-of-speech, syntactical tree and dependency tuple, semantic relations, etc (Zhong et al 2016;Chen et al 2012;Nobata et al 2016).…”

Section: Related Workmentioning

confidence: 99%

“…Despite the fact that most of hate speech resources have been proposed for English, the stateof-the-art for hate speech detection have also proposed automated strategies to expand data for low-resourced languages (Röttger et al 2022). Furthermore, the most relevant literature related to hate speech detection has focused on different tasks, such as (i) automatically detecting hate speech groups such as racism (Hasanuzzaman et al 2017), antisemitism (Ozalp et al 2020;Zannettou et al 2020), religious intolerance (Ghosh Chowdhury et al 2019), misogyny and sexism (Guest et al 2021;Jha and Mamidi 2017), and cyberbullying (Van Hee et al 2015a); (ii) filtering pages with hate and violence (Liu and Forss 2015); (iii) offensive language detection (Vargas et al 2021;Zampieri et al 2019;Steimel et al 2019); (iv) toxic comment detection (Guimarães et al 2020), (v) hateful multi-modal content (Cao et al 2022), and (vi) countering hate speech in dialogue systems (Bonaldi et al 2022). Comprehensive surveys on automatic detection of hate speech in the text were also proposed (Fortuna and Nunes 2018;Poletto et al 2021;Vidgen and Derczynski 2021;Schmidt and Wiegand 2017).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Contextual-Aware and Expert Data Resources for Brazilian Portuguese Hate Speech Detection

Vargas

Carvalho

Pardo

et al. 2023

Preprint

View full text Add to dashboard Cite

This paper provides data resources for low-resource hate speech detection. Specifically, we introduce a large-scale expert annotated corpus of Brazilian Instagram comments and a context-aware offensive lexicon, which was manually extracted by a linguist from the proposed corpus and annotated with contextual information. We further provide native-speaker translations and adaptations from the specialized lexicon for other low-resource languages. The corpus consists of 7,000 document-level multi-layer annotations: (i) a binary offensive class, (ii) offensiveness-level classes, and (iii) nine hate speech targets. The context-aware offensive lexicon holds 1,000 explicit and implicit terms and expressions with pejorative connotations annotated with context-dependent offensiveness and context-independent offensiveness labels. Both corpus and lexicon were annotated by three different experts and achieved high inter-annotator agreement. Finally, we implemented baseline experiments on both data resources (corpus and lexicon). Results show the reliability of the proposed data, outperforming baseline dataset results in Portuguese, as well as presenting promising results for hate speech detection in different languages.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Contextual-Aware and Expert Data Resources for Brazilian Portuguese Hate Speech Detection

Vargas

Carvalho

Pardo

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…4. Lexical resources: In this approach, a controlled vocabulary such as hateful and offensive words and expressions are used as features [18,32,38]. 5.…”

Section: Word Generalizationmentioning

confidence: 99%

“…6. Knowledge-based features: As hate speech detection task is highly contextual-dependent, in this approach, contextual and world knowledge information are used as features [18,40,41]. 7.…”

Section: Word Generalizationmentioning

confidence: 99%

“…The state-of-the-art for offensive language and hate speech detection has focused on different tasks, such as (i) automatically detecting hate speech groups as racism [8], antisemitism [5,13], religious intolerance [14], misogyny and sexism [9,15], and cyberbullying [16]; (ii) filtering pages with hate and violence [17]; (iii) offensive language detection [2,7,18]; and (iv) toxic comment detection [19]. Comprehensive surveys on automatic detection of hate speech in text were proposed by [3,10,12,20].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Contextual-Aware and Expert Resources for Brazilian Portuguese Hate Speech Detection

Vargas

Carvalho

Góes

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

This paper provides the first large-scale expert annotated corpus of Brazilian Instagram comments, a multilayer annotation schema for hate speech and offensive language detection on social media, and a contextual-aware offensive lexicon annotated with contextual information. The HateBR corpus was collected from the comment section of Brazilian politicians' accounts on Instagram and manually annotated by specialists. It is composed of 7,000 documents annotated according to three different layers: a binary classification (offensive versus non-offensive comments), offensiveness-level classification (highly, moderately, and slightly), and nine hate speech groups (xenophobia, racism, homophobia, sexism, religious intolerance, partyism, apology for the dictatorship, antisemitism, and fatphobia). The proposed specialized lexicon was manually identified by a linguist from the proposed corpus, which holds 1,000 explicit and implicit pejorative terms and expressions annotated with contextual information. Both the corpus and the contextual-aware offensive lexicon were annotated by three different experts and achieved high inter-annotator agreement. Lastly, we implemented baseline experiments on our corpus and lexicon. The obtained results outperform the current state-of-the-art for the Portuguese language, and the models which embody our specialized lexicon also present relevant performance in different languages. Accordingly, we hope that the proposed resources foster research on hate speech and offensive language detection in the Natural Language Processing area.

show abstract