SQuARe: A Large-Scale Dataset of Sensitive Questions and Acceptable Responses Created through Human-Machine Collaboration (original) (raw)
Related papers
Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)
Undermining the impact of hateful content with informed and non-aggressive responses, called counter narratives, has emerged as a possible solution for having healthier online communities. Thus, some NLP studies have started addressing the task of counter narrative generation. Although such studies have made an effort to build hate speech / counter narrative (HS/CN) datasets for neural generation, they fall short in reaching either highquality and/or high-quantity. In this paper, we propose a novel human-in-the-loop data collection methodology in which a generative language model is refined iteratively by using its own data from the previous loops to generate new training samples that experts review and/or post-edit. Our experiments comprised several loops including dynamic variations. Results show that the methodology is scalable and facilitates diverse, novel, and cost-effective data collection. To our knowledge, the resulting dataset is the only expertbased multi-target HS/CN dataset available to the community.
Detecting Offensive Content in Open-domain Conversations using Two Stage Semi-supervision
2018
As open-ended human-chatbot interaction becomes commonplace, sensitive content detection gains importance. In this work, we propose a two stage semi-supervised approach to bootstrap large-scale data for automatic sensitive language detection from publicly available web resources. We explore various data selection methods including 1) using a blacklist to rank online discussion forums by the level of their sensitiveness followed by randomly sampling utterances and 2) training a weakly supervised model in conjunction with the blacklist for scoring sentences from online discussion forums to curate a dataset. Our data collection strategy is flexible and allows the models to detect implicit sensitive content for which manual annotations may be difficult. We train models using publicly available annotated datasets as well as using the proposed large-scale semi-supervised datasets. We evaluate the performance of all the models on Twitter and Toxic Wikipedia comments testsets as well as on ...
Agreeing to Disagree: Annotating Offensive Language Datasets with Annotators’ Disagreement
Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, 2021
Since state-of-the-art approaches to offensive language detection rely on supervised learning, it is crucial to quickly adapt them to the continuously evolving scenario of social media. While several approaches have been proposed to tackle the problem from an algorithmic perspective, so to reduce the need for annotated data, less attention has been paid to the quality of these data. Following a trend that has emerged recently, we focus on the level of agreement among annotators while selecting data to create offensive language datasets, a task involving a high level of subjectivity. Our study comprises the creation of three novel datasets of English tweets covering different topics and having five crowd-sourced judgments each. We also present an extensive set of experiments showing that selecting training and test data according to different levels of annotators' agreement has a strong effect on classifiers performance and robustness. Our findings are further validated in cross-domain experiments and studied using a popular benchmark dataset. We show that such hard cases, where low agreement is present, are not necessarily due to poor-quality annotation and we advocate for a higher presence of ambiguous cases in future datasets, particularly in test sets, to better account for the different points of view expressed online.
Ruddit: Norms of Offensiveness for English Reddit Comments
Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)
Warning: This paper contains comments that may be offensive or upsetting. On social media platforms, hateful and offensive language negatively impact the mental well-being of users and the participation of people from diverse backgrounds. Automatic methods to detect offensive language have largely relied on datasets with categorical labels. However, comments can vary in their degree of offensiveness. We create the first dataset of English language Reddit comments that has fine-grained, real-valued scores between-1 (maximally supportive) and 1 (maximally offensive). The dataset was annotated using Best-Worst Scaling, a form of comparative annotation that has been shown to alleviate known biases of using rating scales. We show that the method produces highly reliable offensiveness scores. Finally, we evaluate the ability of widely-used neural models to predict offensiveness scores on this new dataset.
PROVOKE: Toxicity trigger detection in conversations from the top 100 subreddits
PROVOKE: Toxicity trigger detection in conversations from the top 100 subreddits, 2022
Promoting healthy discourse on community-based online platforms like Reddit can be challenging, especially when conversations show ominous signs of toxicity. Therefore, in this study, we find the turning points (i.e., toxicity triggers) making conversations toxic. Before finding toxicity triggers, we built and evaluated various machine learning models to detect toxicity from Reddit comments. Subsequently, we used our best-performing model, a fine-tuned Bidirectional Encoder Representations from Transformers (BERT) model that achieved an area under the receiver operating characteristic curve (AUC) score of 0.983 to detect toxicity. Next, we constructed conversation threads and used the toxicity prediction results to build a training set for detecting toxicity triggers. This procedure entailed using our large-scale dataset to refine toxicity triggers' definition and build a trigger detection dataset using 991,806 conversation threads from the top 100 communities on Reddit. Then, we extracted a set of sentiment shift, topical shift, and context-based features from the trigger detection dataset, using them to build a dual embedding biLSTM neural network that achieved an AUC score of 0.789. Our trigger detection dataset analysis showed that specific triggering keywords are common across all communities, like 'racist' and 'women'. In contrast, other triggering keywords are specific to certain communities, like 'overwatch' in r/Games. Implications are that toxicity trigger detection algorithms can leverage generic approaches but must also tailor detections to specific communities.
RoCHI - International Conference on Human-Computer Interaction
The use of offensive language can lead to uncomfortable situations, psychological harm, and in particular cases even to violence. Social networks and websites struggle to reduce the prevalence of these types of messages by using an automated detector. In this paper, we propose a novel Romanian language dataset for offensive message detection. We manually annotated 4,052 comments on a Romanian local news website into one of the following classes: non-offensive, targeted insults, racist, homophobic, and sexist. In addition, we establish a baseline of five automated classifiers, out of which the model based on RoBERT and two layers of CNN achieves the highest performance with an average F1-score of .74.
Deep learning for detecting inappropriate content in text
International Journal of Data Science and Analytics, 2017
Today, there are a large number of online discussion fora on the internet which are meant for users to express, discuss and exchange their views and opinions on various topics. For example, news portals, blogs, social media channels such as youtube. typically allow users to express their views through comments. In such fora, it has been often observed that user conversations sometimes quickly derail and become inappropriate such as hurling abuses, passing rude and discourteous comments on individuals or certain groups/communities. Similarly, some virtual agents or bots have also been found to respond back to users with inappropriate messages. As a result, inappropriate messages or comments are turning into an online menace slowly degrading the effectiveness of user experiences. Hence, automatic detection and filtering of such inappropriate language has become an important problem for improving the quality of conversations with users as well as virtual agents. In this paper, we propo...
Are These Comments Triggering? Predicting Triggers of Toxicity in Online Discussions
Are These Comments Triggering? Predicting Triggers of Toxicity in Online Discussions, 2020
Understanding the causes or triggers of toxicity adds a new dimension to the prevention of toxic behavior in online discussions. In this research, we define toxicity triggers in online discussions as a non-toxic comment that lead to toxic replies. Then, we build a neural network-based prediction model for toxicity trigger. The prediction model incorporates text-based features and derived features from previous studies that pertain to shifts in sentiment, topic flow, and discussion context. Our findings show that triggers of toxicity contain identifiable features and that incorporating shift features with the discussion context can be detected with a ROC-AUC score of 0.87. We discuss implications for online communities and also possible further analysis of online toxicity and its root causes.
2020
The field of the automatic detection of hate speech and related concepts has raised a lot of interest in the last years. Different datasets were annotated and classified by means of applying different machine learning algorithms. However, few efforts were done in order to clarify the applied categories and homogenize different datasets. Our study takes up this demand. We analyze six different publicly available datasets in this field with respect to their similarity and compatibility. We conduct two different experiments. First, we try to make the datasets compatible and represent the dataset classes as Fast Text word vectors analyzing the similarity between different classes in a intra and inter dataset manner. Second, we submit the chosen datasets to the Perspective API Toxicity classifier, achieving different performances depending on the categories and datasets. One of the main conclusions of these experiments is that many different definitions are being used for equivalent conc...
ETHOS: an Online Hate Speech Detection Dataset
2020
Online hate speech is a newborn problem in our modern society which is growing at a steady rate exploiting weaknesses of the corresponding regimes that characterise several social media platforms. Therefore, this phenomenon is mainly cultivated through such comments, either during users' interaction or on posted multimedia context. Nowadays, giant companies own platforms where many millions of users log in daily. Thus, protection of their users from exposure to similar phenomena for keeping up with the corresponding law, as well as for retaining a high quality of offered services, seems mandatory. Having a robust and reliable mechanism for identifying and preventing the uploading of related material would have a huge effect on our society regarding several aspects of our daily life. On the other hand, its absence would deteriorate heavily the total user experience, while its erroneous operation might raise several ethical issues. In this work, we present a protocol for creating ...