Workplace communication (e.g. email, chat, etc.) is a central part of enterprise productivity. Healthy conversations are crucial for creating an inclusive environment and maintaining harmony in an organization. Toxic communications at the workplace c
an negatively impact overall job satisfaction and are often subtle, hidden, or demonstrate human biases. The linguistic subtlety of mild yet hurtful conversations has made it difficult for researchers to quantify and extract toxic conversations automatically. While offensive language or hate speech has been extensively studied in social communities, there has been little work studying toxic communication in emails. Specifically, the lack of corpus, sparsity of toxicity in enterprise emails, and well-defined criteria for annotating toxic conversations have prevented researchers from addressing the problem at scale. We take the first step towards studying toxicity in workplace emails by providing (1) a general and computationally viable taxonomy to study toxic language at the workplace (2) a dataset to study toxic language at the workplace based on the taxonomy and (3) analysis on why offensive language and hate-speech datasets are not suitable to detect workplace toxicity.
Toxic comments contain forms of non-acceptable language targeted towards groups or individuals. These types of comments become a serious concern for government organizations, online communities, and social media platforms. Although there are some app
roaches to handle non-acceptable language, most of them focus on supervised learning and the English language. In this paper, we deal with toxic comment detection as a semi-supervised strategy over a heterogeneous graph. We evaluate the approach on a toxic dataset of the Portuguese language, outperforming several graph-based methods and achieving competitive results compared to transformer architectures.
The availability of language representations learned by large pretrained neural network models (such as BERT and ELECTRA) has led to improvements in many downstream Natural Language Processing tasks in recent years. Pretrained models usually differ i
n pretraining objectives, architectures, and datasets they are trained on which can affect downstream performance. In this contribution, we fine-tuned German BERT and German ELECTRA models to identify toxic (subtask 1), engaging (subtask 2), and fact-claiming comments (subtask 3) in Facebook data provided by the GermEval 2021 competition. We created ensembles of these models and investigated whether and how classification performance depends on the number of ensemble members and their composition. On out-of-sample data, our best ensemble achieved a macro-F1 score of 0.73 (for all subtasks), and F1 scores of 0.72, 0.70, and 0.76 for subtasks 1, 2, and 3, respectively.
This paper describes our methods submitted for the GermEval 2021 shared task on identifying toxic, engaging and fact-claiming comments in social media texts (Risch et al., 2021). We explore simple strategies for semi-automatic generation of rule-base
d systems with high precision and low recall, and use them to achieve slight overall improvements over a standard BERT-based classifier.
In this paper, we report on our approach to addressing the GermEval 2021 Shared Task on the Identification of Toxic, Engaging, and Fact-Claiming Comments for the German language. We submitted three runs for each subtask based on ensembles of three mo
dels each using contextual embeddings from pre-trained language models using SVM and neural-network-based classifiers. We include language-specific as well as language-agnostic language models -- both with and without fine-tuning. We observe that for the runs we submitted that the SVM models overfitted the training data and this affected the aggregation method (simple majority voting) of the ensembles. The model records a lower performance on the test set than on the training set. Exploring the issue of overfitting we uncovered that due to a bug in the pipeline the runs we submitted had not been trained on the full set but only on a small training set. Therefore in this paper we also include the results we get when trained on the full training set which demonstrate the power of ensembles.
In this work, we present our approaches on the toxic comment classification task (subtask 1) of the GermEval 2021 Shared Task. For this binary task, we propose three models: a German BERT transformer model; a multilayer perceptron, which was first tr
ained in parallel on textual input and 14 additional linguistic features and then concatenated in an additional layer; and a multilayer perceptron with both feature types as input. We enhanced our pre-trained transformer model by re-training it with over 1 million tweets and fine-tuned it on two additional German datasets of similar tasks. The embeddings of the final fine-tuned German BERT were taken as the textual input features for our neural networks. Our best models on the validation data were both neural networks, however our enhanced German BERT gained with a F1-score = 0.5895 a higher prediction on the test data.
With the rapid growth in technology, social media activity has seen a boom across all age groups. It is humanly impossible to check all the tweets, comments and status manually whether they follow proper community guidelines. A lot of toxicity is reg
ularly posted on these social media platforms. This research aims to find toxic words in a sentence so that a healthy social community is built across the globe and the users receive censored content with specific warnings and facts. To solve this challenging problem, authors have combined concepts of Linked List for pre-processing and then used the idea of stacked embeddings like BERT Embeddings, Flair Embeddings and Word2Vec on the flairNLP framework to get the desired results. F1 metric was used to evaluate the model. The authors were able to produce a 0.74 F1 score on their test set.
With the rise of research on toxic comment classification, more and more annotated datasets have been released. The wide variety of the task (different languages, different labeling processes and schemes) has led to a large amount of heterogeneous da
tasets that can be used for training and testing very specific settings. Despite recent efforts to create web pages that provide an overview, most publications still use only a single dataset. They are not stored in one central database, they come in many different data formats and it is difficult to interpret their class labels and how to reuse these labels in other projects. To overcome these issues, we present a collection of more than thirty datasets in the form of a software tool that automatizes downloading and processing of the data and presents them in a unified data format that also offers a mapping of compatible class labels. Another advantage of that tool is that it gives an overview of properties of available datasets, such as different languages, platforms, and class labels to make it easier to select suitable training and test data.
The Toxic Spans Detection task of SemEval-2021 required participants to predict the spans of toxic posts that were responsible for the toxic label of the posts. The task could be addressed as supervised sequence labeling, using training data with gol
d toxic spans provided by the organisers. It could also be treated as rationale extraction, using classifiers trained on potentially larger external datasets of posts manually annotated as toxic or not, without toxic span annotations. For the supervised sequence labeling approach and evaluation purposes, posts previously labeled as toxic were crowd-annotated for toxic spans. Participants submitted their predicted spans for a held-out test set and were scored using character-based F1. This overview summarises the work of the 36 teams that provided system descriptions.
In this work, we present our approach and findings for SemEval-2021 Task 5 - Toxic Spans Detection. The task's main aim was to identify spans to which a given text's toxicity could be attributed. The task is challenging mainly due to two constraints:
the small training dataset and imbalanced class distribution. Our paper investigates two techniques, semi-supervised learning and learning with Self-Adjusting Dice Loss, for tackling these challenges. Our submitted system (ranked ninth on the leader board) consisted of an ensemble of various pre-trained Transformer Language Models trained using either of the above-proposed techniques.
