NLP research has attained high performances in abusive language detection as a supervised classification task. While in research settings, training and test datasets are usually obtained from similar data samples, in practice systems are often applied on data that are different from the training set in topic and class distributions. Also, the ambiguity in class definitions inherited in this task aggravates the discrepancies between source and target datasets. We explore the topic bias and the task formulation bias in cross-dataset generalization. We show that the benign examples in the Wikipedia Detox dataset are biased towards platform-specific topics. We identify these examples using unsupervised topic modeling and manual inspection of topics' keywords. Removing these topics increases cross-dataset generalization, without reducing in-domain classification performance. For a robust dataset design, we suggest applying inexpensive unsupervised methods to inspect the collected data and downsize the non-generalizable content before manually annotating for class labels.
One challenge that social media platforms are facing nowadays is hate speech. Hence, automatic hate speech detection has been increasingly researched in recent years - in particular with the rise of deep learning. A problem of these models is their vulnerability to undesirable bias in training data. We investigate the impact of political bias on hate speech classification by constructing three politically-biased data sets (left-wing, right-wing, politically neutral) and compare the performance of classifiers trained on them. We show that (1) political bias negatively impairs the performance of hate speech classifiers and (2) an explainable machine learning model can help to visualize such bias within the training data. The results show that political bias in training data has an impact on hate speech classification and can become a serious issue.
A challenge that many online platforms face is hate speech or any other form of online abuse. To cope with this, hate speech detection systems are developed based on machine learning to reduce manual work for monitoring these platforms. Unfortunately, machine learning is vulnerable to unintended bias in training data, which could have severe consequences, such as a decrease in classification performance or unfair behavior (e.g., discriminating minorities). In the scope of this study, we want to investigate annotator bias — a form of bias that annotators cause due to different knowledge in regards to the task and their subjective perception. Our goal is to identify annotation bias based on similarities in the annotation behavior from annotators. To do so, we build a graph based on the annotations from the different annotators, apply a community detection algorithm to group the annotators, and train for each group classifiers whose performances we compare. By doing so, we are able to identify annotator bias within a data set. The proposed method and collected insights can contribute to developing fairer and more reliable hate speech classification models.
Automated detection of abusive language online has become imperative. Current sequential models (LSTM) do not work well for long and complex sentences while bi-transformer models (BERT) are not computationally efficient for the task. We show that classifiers based on syntactic structure of the text, dependency graphical convolutional networks (DepGCNs) can achieve state-of-the-art performance on abusive language datasets. The overall performance is at par with of strong baselines such as fine-tuned BERT. Further, our GCN-based approach is much more efficient than BERT at inference time making it suitable for real-time detection.
Recently, a few studies have discussed the limitations of datasets collected for the task of detecting hate speech from different viewpoints. We intend to contribute to the conversation by providing a consolidated overview of these issues pertaining to the data that debilitate research in this area. Specifically, we discuss how the varying pre-processing steps and the format for making data publicly available result in highly varying datasets that make an objective comparison between studies difficult and unfair. There is currently no study (to the best of our knowledge) focused on comparing the attributes of existing datasets for hate speech detection, outlining their limitations and recommending approaches for future research. This work intends to fill that gap and become the one-stop shop for information regarding hate speech datasets.
