STUFIIT at SemEval-2019 Task 5: Multilingual Hate Speech Detection on Twitter with MUSE and ELMo Embeddings

Bojkovský, Michal; Pikuliak, Matúš

doi:10.18653/v1/s19-2082

Cited by 18 publications

(15 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The deep learning methods can be roughly divided into two categories: one focuses on front-end processing which optimizes the word embedding technology, and the other on mid-end processing which usually uses simple word or character based embedding technology and pays more attention to the middle neural networks processing. The most famous methods focused on front-end processing are Embeddings from Language Models (ELMo) [6] [13], which trains word vectors with context, and Bidirectional Encoder Representation from Transformers (BERT) [14] [15]. BERT is the first deeply bidirectional, unsupervised language representation from unlabeled text by jointly conditioning on both left and right context in all layers.…”

Section: B State Of the Art In Deep Learningmentioning

confidence: 99%

“…Recently, machine learning approach which can learn the different associations between pieces of text, and that a particular output is expected for a particular input by using pre-labeled examples as training data is popular in scientific studies for hate speech detection. Among various machine learning methods, deep learning which is a subset of machine learning, is very prominent in Natural Language Processing (NLP) to tackle the issue of text classification [5] [6].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Deep Learning Based Fusion Approach for Hate Speech Detection

et al. 2020

View full text Add to dashboard Cite

In recent years, the increasing prevalence of hate speech in social media has been considered as a serious problem worldwide. Many governments and organizations have made significant investment in hate speech detection techniques, which have also attracted the attention of the scientific community. Although plenty of literature focusing on this issue is available, it remains difficult to assess the performances of each proposed method, as each has its own advantages and disadvantages. A general way to improve the overall results of classification by fusing the various classifiers results is a meaningful attempt. We first focus on several famous machine learning methods for text classification such as Embeddings from Language Models (ELMo), Bidirectional Encoder Representation from Transformers (BERT) and Convolutional Neural Network (CNN), and apply these methods to the data sets of the SemEval 2019 Task 5. We then adopt some fusion strategies to combine the classifiers to improve the overall classification performance. The results show that the accuracy and F1-score of the classification are significantly improved.

show abstract

Section: B State Of the Art In Deep Learningmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Deep Learning Based Fusion Approach for Hate Speech Detection

et al. 2020

View full text Add to dashboard Cite

show abstract

“…Global Vectors for word representation (GloVe) [6] and random embeddings as input to DNN classifiers has been compared in [7]. Recently, sentence embeddings [9] and Embeddings from Language Models (ELMo) [10] were used as input to classifiers for toxic comment classification. Multi-features based approach combining various lexicons and semantic-based features is presented in [11].…”

Section: She Looks Like a Plastic Monkey Doll!mentioning

confidence: 99%

BERT and fastText Embeddings for Automatic Detection of Toxic Speech

D'Sa

Illina

Fohr

2020

2020 International Multi-Conference On: “Organization of Knowledge and Advanced Technologies” (OCTA)

View full text Add to dashboard Cite

With the expansion of Internet usage, catering to the dissemination of thoughts and expressions of an individual, there has been an immense increase in the spread of online hate speech. Social media, community forums, discussion platforms are few examples of common playground of online discussions where people are freely allowed to communicate. However, the freedom of speech may be misused by some people by arguing aggressively, offending others and spreading verbal violence. As there is no clear distinction between the terms offensive, abusive, hate and toxic speech, in this paper we consider the above mentioned terms as toxic speech. In many countries, online toxic speech is punishable by the law. Thus, it is important to automatically detect and remove toxic speech from online medias. Through this work, we propose automatic classification of toxic speech using embedding representations of words and deep-learning techniques. We perform binary and multi-class classification using a Twitter corpus and study two approaches: (a) a method which consists in extracting of word embeddings and then using a DNN classifier; (b) fine-tuning the pre-trained BERT model. We observed that BERT fine-tuning performed much better. Proposed methodology can be used for any other type of social media comments.

show abstract

“…To protect these users from online harassment, it is necessary to develop a tool that can automatically detect the toxic language in social media. In fact, many toxic language detection (TLD) systems have been proposed in these years based on different models, such as support vector machines (SVM) (Gaydhani et al, 2018), bi-directional long shortterm memory (BiLSTM) (Bojkovskỳ and Pikuliak, 2019), logistic regression and fine-tuning BERT (d'Sa et al, 2020).…”

Section: Introductionmentioning

confidence: 99%

Proceedings of the 5th Workshop on Online Abuse and Harms (WOAH 2021)

2021

View full text Add to dashboard Cite

Message from the OrganisersDigital technologies have brought myriad benefits for society, transforming how people connect, communicate and interact with each other. However, they have also enabled harmful and abusive behaviours to reach large audiences and for their negative effects to be amplified, including interpersonal aggression, bullying and hate speech. Already marginalised and vulnerable communities are often disproportionately at risk of receiving such abuse, compounding other social inequalities and injustices. The Workshop on Online Abuse and Harms (WOAH) convenes research into these issues, particularly work that develops, interrogates and applies computational methods for detecting, classifying and modelling online abuse.Technical disciplines such as machine learning and natural language processing (NLP) have made substantial advances in creating more powerful technologies to stop online abuse. Yet a growing body of work shows the limitations of many automated detection systems for tackling abusive online content, which can be biased, brittle, low performing and simplistic. These issues are magnified by the lack of explainability and transparency. And although WOAH is collocated with ACL and many of our papers are rooted firmly in the field of machine learning, these are not purely engineering challenges, but raise fundamental social questions of fairness and harm. For this reason, we continue to emphasise the need for inter-, cross-and anti-disciplinary work by inviting contributions from a range of fields, including but not limited to: NLP, machine learning, computational social sciences, law, politics, psychology, network analysis, sociology and cultural studies. In this fifth edition of WOAH we direct the conversation at the workshop through our theme: Social Bias and Unfairness in Online Abuse Detection Systems. Continuing the tradition started in WOAH 4, we have invited civil society, in particular individuals and organisations working with women and marginalised communities, to submit reports, case studies, findings, data, and to record their lived experiences through our civil society track. Our hope is that WOAH provides a platform to facilitate the interdisciplinary conversations and collaborations that are needed to effectively and ethically address online abuse.Speaking to the complex nature of the issue of online abuse, we are pleased to invite Leon Derczynski, currently an Associate Professor at ITU Copenhagen who works on a range of topics in Natural Language Processing; Deb Raji, currently a Research Fellow at Mozilla who researches AI accountability and auditing; Murali Shanmugavelan, currently a researcher at the Centre for Global Media and Communications at SOAS (London) to deliver keynotes. We are grateful to all our speakers for being available, and look forward to the dialogues that they will generate. On the day of WOAH the invited keynote speakers will give talks and then take part in a multi-disciplinary panel discussion to debate our theme and other issues in computational o...

show abstract

STUFIIT at SemEval-2019 Task 5: Multilingual Hate Speech Detection on Twitter with MUSE and ELMo Embeddings

Cited by 18 publications

References 9 publications

Deep Learning Based Fusion Approach for Hate Speech Detection

Deep Learning Based Fusion Approach for Hate Speech Detection

BERT and fastText Embeddings for Automatic Detection of Toxic Speech

Proceedings of the 5th Workshop on Online Abuse and Harms (WOAH 2021)

Contact Info

Product

Resources

About