In "Offensive keyword extraction based on the attention mechanism of BERT and the eigenvector centrality using a graph representation," published in Personal and Ubiquitous Computing 27 (1), 45-57, an oversight occurred regarding the clarity and comprehensibility of the proposed method.This erratum serves to rectify and clarify the methodological description to enhance readers' understanding of the research.

Cross-lingual transfer learning from highresource to medium and low-resource languages has shown encouraging results.However, the scarcity of resources in target languages remains a challenge.In this work, we resort to data augmentation and continual pre-training for domain adaptation to improve cross-lingual abusive language detection.For data augmentation, we analyze two existing techniques based on vicinal risk minimization and propose MIXAG, a novel data augmentation method which interpolates pairs of instances based on the angle of their representations.Our experiments involve seven languages typologically distinct from English and three different domains.The results reveal that the data augmentation strategies can enhance fewshot cross-lingual abusive language detection.Specifically, we observe that consistently in all target languages, MIXAG improves significantly in multidomain and multilingual environments.Finally, we show through an error analysis how the domain adaptation can favour the class of abusive texts (reducing false negatives), but at the same time, declines the precision of the abusive language detection model.

Hate speech detection refers broadly to the automatic identification of language that may be considered discriminatory against certain groups of people. The goal is to help online platforms to identify and remove harmful content. Humans are usually capable of detecting hatred in critical cases, such as when the hatred is non-explicit, but how do computer models address this situation? In this work, we aim to contribute to the understanding of ethical issues related to hate speech by analysing two transformer-based models trained to detect hate speech. Our study focuses on analysing the relationship between these models and a set of hateful keywords extracted from the three well-known datasets. For the extraction of the keywords, we propose a metric that takes into account the division among classes to favour the most common words in hateful contexts. In our experiments, we first compared the overlap between the extracted keywords with the words to which the models pay the most attention in decision-making. On the other hand, we investigate the bias of the models towards the extracted keywords. For the bias analysis, we characterize and use two metrics and evaluate two strategies to try to mitigate the bias. Surprisingly, we show that over 50% of the salient words of the models are not hateful and that there is a higher number of hateful words among the extracted keywords. However, we show that the models appear to be biased towards the extracted keywords. Experimental results suggest that fitting models with hateful texts that do not contain any of the keywords can reduce bias and improve the performance of the models.

Abstract The proliferation of harmful content on social media affects a large part of the user community. Therefore, several approaches have emerged to control this phenomenon automatically. However, this is still a quite challenging task. In this paper, we explore the offensive language as a particular case of harmful content and focus our study in the analysis of keywords in available datasets composed of offensive tweets. Thus, we aim to identify relevant words in those datasets and analyze how they can affect model learning. For keyword extraction, we propose an unsupervised hybrid approach which combines the multi-head self-attention of BERT and a reasoning on a word graph. The attention mechanism allows to capture relationships among words in a context, while a language model is learned. Then, the relationships are used to generate a graph from what we identify the most relevant words by using the eigenvector centrality. Experiments were performed by means of two mechanisms. On the one hand, we used an information retrieval system to evaluate the impact of the keywords in recovering offensive tweets from a dataset. On the other hand, we evaluated a keyword-based model for offensive language detection. Results highlight some points to consider when training models with available datasets.

Hate speech detection is a prominent and challenging task, since hate messages are often expressed in subtle ways and with characteristics that may vary depending on the author. Hence, many models suffer from the generalization problem. However, retrieving and monitoring hateful content on social media is a current necessity. In this paper, we propose an unsupervised approach using Graph Auto-Encoders (GAE), which allows us to avoid using labeled data when training the representation of the texts. Specifically, we represent texts as nodes of a graph, and use a transformer layer together with a convolutional layer to encode these nodes in a low-dimensional space. As a result, we obtain embeddings that can be decoded into a reconstruction of the original network. Our main idea is to learn a model with a set of texts without supervision, in order to generate embeddings for the nodes: nodes with the same label should be close in the embedding space, which, in turn, should allow us to distinguish among classes. We employ this strategy to detect hate speech in multi-domain and multilingual sets of texts, where our method shows competitive results on small datasets.

Automatic hate speech detection has become a crucial task nowadays, due the increase of hate on the Internet and its negative consequences. Therefore, in our PhD we propose the design and implementation of methods for the automatic processing of hate messages. The study is focused on the hate messages on Twitter. The hypothesis on which the research is based is that the prediction of hate speech, considering textual content, can be improved by the combination of features such as the activity and communities of users, as well as the images that can be shared with the tweets. In this way, we intend to develop strategies for the automatic detection of hate with multimodal and also multilingual (both in English and Spanish) approaches. Furthermore, our research includes the study of counter-narrative as an alternative to mitigate the effects of hate speech. To address the problem, we employ deep learning techniques, deepening the study of approaches based on representation with graphs.