A carregar...
Publicação
Análise de sentimentos e opinion mining para a análise de discurso de ódio sexista em Português Europeu
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 6.57 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
A presente dissertação analisa o discurso sexista no Português Europeu nas redes sociais Twitter e Facebook, com o objetivo de compreender se este deve ser entendido como discurso de ódio, de que modo se manifesta e pode ser detetado, e se se correlaciona com a análise de sentimentos.
A investigação tem também como objetivos discutir o problema conceptual do ‘discurso de ódio’, sobretudo na ausência de uma definição “universal”, e explorar a forma como este se relaciona com manifestações de sexismo. Além disso, procura demonstrar que, devido à natureza da linguagem, as ferramentas automáticas podem obter resultados que divergem da interpretação humana, sobretudo na análise de sentimentos do discurso sexista.
O estudo combina técnicas de opinion mining para a recolha de publicações potencialmente sexistas com anotação manual, análise de sentimentos, análise de concordância entre anotadores e uso de ferramentas para a extração e observação de padrões linguísticos que caracterizam o discurso em estudo. Esta metodologia originou um corpus anotado manualmente e complementado com a análise de sentimentos, permitindo examinar a presença e o tipo de sexismo, bem como a polaridade de sentimentos, com particular atenção ao público-alvo.
Os resultados revelam a existência de desigualdade e hostilidade predominantemente dirigida às mulheres. Observou-se que os discursos feministas se centram na descrição e denúncia, enquanto os antifeministas tendem a incitar à violência. Por fim, a análise evidencia que fatores como género e faixa etária influenciam a perceção e anotação de conteúdos sexistas, refletindo a complexidade do fenómeno nas redes sociais.
This thesis examines sexist speech in European Portuguese on the social media platform Twitter and Facebook. The study aims to determine whether it should be understood as a form of hate speech, how it manifests, how it can be detected, and how it correlates with sentiment analysis. The research further addresses the conceptual problem of ‘hate speech’ due to the lack of universal definition and explores its relationship to manifestations of sexism. Furthermore, the study seeks to demonstrate that, due to the language’s nature, automatic tools may yield results that diverge from human interpretation, especially in the context of sentiment analysis of sexist discourse. The study combines opinion mining techniques to collect potential sexist posts along manual annotation, sentiment analysis, inter-annotator agreement evaluation, and use of tools to extract and observe linguistic patterns that can characterize the speech under study. This methodology resulted in a manually annotated corpus, complemented by sentiment analysis, allowing for an examination of presence and type of sexism, as well sentiment polarity, with particular attention to the affected target group. The results show the presence of inequality and hostility mainly aimed at women. It was noted that feminist speech tends to focus on description and denunciation, while antifeminist discourse is more likely to incite violence. The analysis also finds that factors such as gender and age shape how sexist content is perceived and annotated, showing the complexity of the phenomenon on social media.
This thesis examines sexist speech in European Portuguese on the social media platform Twitter and Facebook. The study aims to determine whether it should be understood as a form of hate speech, how it manifests, how it can be detected, and how it correlates with sentiment analysis. The research further addresses the conceptual problem of ‘hate speech’ due to the lack of universal definition and explores its relationship to manifestations of sexism. Furthermore, the study seeks to demonstrate that, due to the language’s nature, automatic tools may yield results that diverge from human interpretation, especially in the context of sentiment analysis of sexist discourse. The study combines opinion mining techniques to collect potential sexist posts along manual annotation, sentiment analysis, inter-annotator agreement evaluation, and use of tools to extract and observe linguistic patterns that can characterize the speech under study. This methodology resulted in a manually annotated corpus, complemented by sentiment analysis, allowing for an examination of presence and type of sexism, as well sentiment polarity, with particular attention to the affected target group. The results show the presence of inequality and hostility mainly aimed at women. It was noted that feminist speech tends to focus on description and denunciation, while antifeminist discourse is more likely to incite violence. The analysis also finds that factors such as gender and age shape how sexist content is perceived and annotated, showing the complexity of the phenomenon on social media.
Descrição
Palavras-chave
Discurso sexista Discurso de ódio Sexismo Opinion mining Análise de sentimentos Corpus Redes sociais Discurso feminista Discurso antifeminista Sexist speech Hate speech Sexism Sentiment analysis Social media Feminist speech Antifeminist speech