Machine learning-based classifiers have been used for text classification, such as sentiment analysis, news classification, and toxic comment classification. However, supervised machine learning models often require large amounts of labeled data for training, and manual annotation is both labor-intensive and requires domain-specific knowledge, leading to relatively high annotation costs. To address this issue, we propose an approach that integrates large language models (LLMs) into an active learning framework, achieving high cross-task text classification performance without the need for any manually labeled data. Furthermore, compared to directly applying GPT for classification tasks, our approach retains over 93% of its classification performance while requiring only approximately 6% of the computational time and monetary cost, effectively balancing performance and resource efficiency. These findings provide new insights into the efficient utilization of LLMs and active learning algorithms in text classification tasks, paving the way for their broader application.
AI regulations are expected to prohibit machine learning models from using sensitive attributes during training. However, the latest Natural Language Processing (NLP) classifiers, which rely on deep learning, operate as black-box systems, complicating the detection and remediation of such misuse. Traditional bias mitigation methods in NLP aim for comparable performance across different groups based on attributes like gender or race but fail to address the underlying issue of reliance on protected attributes. To partly fix that, we introduce NLPGuard, a framework for mitigating the reliance on protected attributes in NLP classifiers. NLPGuard takes an unlabeled dataset, an existing NLP classifier, and its training data as input, producing a modified training dataset that significantly reduces dependence on protected attributes without compromising accuracy. NLPGuard is applied to three classification tasks: identifying toxic language, sentiment analysis, and occupation classification. Our evaluation shows that current NLP classifiers heavily depend on protected attributes, with up to $23\%$ of the most predictive words associated with these attributes. However, NLPGuard effectively reduces this reliance by up to $79\%$, while slightly improving accuracy.
We present PyTorch Frame, a PyTorch-based framework for deep learning over multi-modal tabular data. PyTorch Frame makes tabular deep learning easy by providing a PyTorch-based data structure to handle complex tabular data, introducing a model abstraction to enable modular implementation of tabular models, and allowing external foundation models to be incorporated to handle complex columns (e.g., LLMs for text columns). We demonstrate the usefulness of PyTorch Frame by implementing diverse tabular models in a modular way, successfully applying these models to complex multi-modal tabular data, and integrating our framework with PyTorch Geometric, a PyTorch library for Graph Neural Networks (GNNs), to perform end-to-end learning over relational databases.
Large language models (LLMs) have attracted considerable interest in the fields of natural language understanding (NLU) and natural language generation (NLG) since their introduction. In contrast, the legacy of Capsule Neural Networks (CapsNet) appears to have been largely forgotten amidst all of this excitement. This project's objective is to reignite interest in CapsNet by reopening the previously closed studies and conducting a new research into CapsNet's potential. We present a study where CapsNet is used to classify toxic text by leveraging pre-trained BERT embed dings (bert-base-uncased) on a large multilingual dataset. In this experiment, CapsNet was tasked with categorizing toxic text. By comparing the performance of CapsNet to that of other architectures, such as DistilBERT, Vanilla Neural Networks (VNN), and Convolutional Neural Networks (CNN), we were able to achieve an accuracy of 90.44 %. This result highlights the benefits of CapsNet over text data and suggests new ways to enhance their performance so that it is comparable to DistilBERT and other reduced architectures.
Recent developments in large language models (LLMs) have been impressive. However, these models sometimes show inconsistencies and problematic behavior, such as hallucinating facts, generating flawed code, or creating offensive and toxic content. Unlike these models, humans typically utilize external tools to cross-check and refine their initial content, like using a search engine for fact-checking, or a code interpreter for debugging. Inspired by this observation, we introduce a framework called CRITIC that allows LLMs, which are essentially "black boxes" to validate and progressively amend their own outputs in a manner similar to human interaction with tools. More specifically, starting with an initial output, CRITIC interacts with appropriate tools to evaluate certain aspects of the text, and then revises the output based on the feedback obtained during this validation process. Comprehensive evaluations involving free-form question answering, mathematical program synthesis, and toxicity reduction demonstrate that CRITIC consistently enhances the performance of LLMs. Meanwhile, our research highlights the crucial importance of external feedback in promoting the ongoing self-improvement of LLMs.
The rapid growth in user generated content on social media has resulted in a significant rise in demand for automated content moderation. Various methods and frameworks have been proposed for the tasks of hate speech detection and toxic comment classification. In this work, we combine common datasets to extend these tasks to brand safety. Brand safety aims to protect commercial branding by identifying contexts where advertisements should not appear and covers not only toxicity, but also other potentially harmful content. As these datasets contain different label sets, we approach the overall problem as a binary classification task. We demonstrate the need for building brand safety specific datasets via the application of common toxicity detection datasets to a subset of brand safety and empirically analyze the effects of weighted sampling strategies in text classification.
Toxic comment detection on social media has proven to be essential for content moderation. This paper compares a wide set of different models on a highly skewed multi-label hate speech dataset. We consider inference time and several metrics to measure performance and bias in our comparison. We show that all BERTs have similar performance regardless of the size, optimizations or language used to pre-train the models. RNNs are much faster at inference than any of the BERT. BiLSTM remains a good compromise between performance and inference time. RoBERTa with Focal Loss offers the best performance on biases and AUROC. However, DistilBERT combines both good AUROC and a low inference time. All models are affected by the bias of associating identities. BERT, RNN, and XLNet are less sensitive than the CNN and Compact Convolutional Transformers.
On the world wide web, toxic content detectors are a crucial line of defense against potentially hateful and offensive messages. As such, building highly effective classifiers that enable a safer internet is an important research area. Moreover, the web is a highly multilingual, cross-cultural community that develops its own lingo over time. As such, it is crucial to develop models that are effective across a diverse range of languages, usages, and styles. In this paper, we present the fundamentals behind the next version of the Perspective API from Google Jigsaw. At the heart of the approach is a single multilingual token-free Charformer model that is applicable across a range of languages, domains, and tasks. We demonstrate that by forgoing static vocabularies, we gain flexibility across a variety of settings. We additionally outline the techniques employed to make such a byte-level model efficient and feasible for productionization. Through extensive experiments on multilingual toxic comment classification benchmarks derived from real API traffic and evaluation on an array of code-switching, covert toxicity, emoji-based hate, human-readable obfuscation, distribution shift, and bias evaluation settings, we show that our proposed approach outperforms strong baselines. Finally, we present our findings from deploying this system in production.
While in real life everyone behaves themselves at least to some extent, it is much more difficult to expect people to behave themselves on the internet, because there are few checks or consequences for posting something toxic to others. Yet, for people on the other side, toxic texts often lead to serious psychological consequences. Detecting such toxic texts is challenging. In this paper, we attempt to build a toxicity detector using machine learning methods including CNN, Naive Bayes model, as well as LSTM. While there has been numerous groundwork laid by others, we aim to build models that provide higher accuracy than the predecessors. We produced very high accuracy models using LSTM and CNN, and compared them to the go-to solutions in language processing, the Naive Bayes model. A word embedding approach is also applied to empower the accuracy of our models.
Sociodemographic biases are a common problem for natural language processing, affecting the fairness and integrity of its applications. Within sentiment analysis, these biases may undermine sentiment predictions for texts that mention personal attributes that unbiased human readers would consider neutral. Such discrimination can have great consequences in the applications of sentiment analysis both in the public and private sectors. For example, incorrect inferences in applications like online abuse and opinion analysis in social media platforms can lead to unwanted ramifications, such as wrongful censoring, towards certain populations. In this paper, we address the discrimination against people with disabilities, PWD, done by sentiment analysis and toxicity classification models. We provide an examination of sentiment and toxicity analysis models to understand in detail how they discriminate PWD. We present the Bias Identification Test in Sentiments (BITS), a corpus of 1,126 sentences designed to probe sentiment analysis models for biases in disability. We use this corpus to demonstrate statistically significant biases in four widely used sentiment analysis tools (TextBlob, VADER, Google Cloud Natural Language API and DistilBERT) and two toxicity analysis models trained to predict toxic comments on Jigsaw challenges (Toxic comment classification and Unintended Bias in Toxic comments). The results show that all exhibit strong negative biases on sentences that mention disability. We publicly release BITS Corpus for others to identify potential biases against disability in any sentiment analysis tools and also to update the corpus to be used as a test for other sociodemographic variables as well.
Understanding toxicity in user conversations is undoubtedly an important problem. Addressing "covert" or implicit cases of toxicity is particularly hard and requires context. Very few previous studies have analysed the influence of conversational context in human perception or in automated detection models. We dive deeper into both these directions. We start by analysing existing contextual datasets and come to the conclusion that toxicity labelling by humans is in general influenced by the conversational structure, polarity and topic of the context. We then propose to bring these findings into computational detection models by introducing and evaluating (a) neural architectures for contextual toxicity detection that are aware of the conversational structure, and (b) data augmentation strategies that can help model contextual toxicity detection. Our results have shown the encouraging potential of neural architectures that are aware of the conversation structure. We have also demonstrated that such models can benefit from synthetic data, especially in the social media domain.
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 in 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.
Toxic comment classification models are often found biased toward identity terms which are terms characterizing a specific group of people such as "Muslim" and "black". Such bias is commonly reflected in false-positive predictions, i.e. non-toxic comments with identity terms. In this work, we propose a novel approach to tackle such bias in toxic comment classification, leveraging the notion of subjectivity level of a comment and the presence of identity terms. We hypothesize that when a comment is made about a group of people that is characterized by an identity term, the likelihood of that comment being toxic is associated with the subjectivity level of the comment, i.e. the extent to which the comment conveys personal feelings and opinions. Building upon the BERT model, we propose a new structure that is able to leverage these features, and thoroughly evaluate our model on 4 datasets of varying sizes and representing different social media platforms. The results show that our model can consistently outperform BERT and a SOTA model devised to address identity term bias in a different way, with a maximum improvement in F1 of 2.43% and 1.91% respectively.
Despite the high accuracy offered by state-of-the-art deep natural-language models (e.g. LSTM, BERT), their application in real-life settings is still widely limited, as they behave like a black-box to the end-user. Hence, explainability is rapidly becoming a fundamental requirement of future-generation data-driven systems based on deep-learning approaches. Several attempts to fulfill the existing gap between accuracy and interpretability have been done. However, robust and specialized xAI (Explainable Artificial Intelligence) solutions tailored to deep natural-language models are still missing. We propose a new framework, named T-EBAnO, which provides innovative prediction-local and class-based model-global explanation strategies tailored to black-box deep natural-language models. Given a deep NLP model and the textual input data, T-EBAnO provides an objective, human-readable, domain-specific assessment of the reasons behind the automatic decision-making process. Specifically, the framework extracts sets of interpretable features mining the inner knowledge of the model. Then, it quantifies the influence of each feature during the prediction process by exploiting the novel normalized Perturbation Influence Relation index at the local level and the novel Global Absolute Influence and Global Relative Influence indexes at the global level. The effectiveness and the quality of the local and global explanations obtained with T-EBAnO are proved on (i) a sentiment analysis task performed by a fine-tuned BERT model, and (ii) a toxic comment classification task performed by an LSTM model.
Customer product reviews play a role in improving the quality of products and services for business organizations or their brands. Complaining is an attitude that expresses dissatisfaction with an event or a product not meeting customer expectations. In this paper, we build a Open-domain Complaint Detection dataset (UIT-ViOCD), including 5,485 human-annotated reviews on four categories about product reviews on e-commerce sites. After the data collection phase, we proceed to the annotation task and achieve the inter-annotator agreement Am of 87%. Then, we present an extensive methodology for the research purposes and achieve 92.16% by F1-score for identifying complaints. With the results, in the future, we aim to build a system for open-domain complaint detection in E-commerce websites.
The rise of social media has led to the increasing of comments on online forums. However, there still exists invalid comments which are not informative for users. Moreover, those comments are also quite toxic and harmful to people. In this paper, we create a dataset for constructive and toxic speech detection, named UIT-ViCTSD (Vietnamese Constructive and Toxic Speech Detection dataset) with 10,000 human-annotated comments. For these tasks, we propose a system for constructive and toxic speech detection with the state-of-the-art transfer learning model in Vietnamese NLP as PhoBERT. With this system, we obtain F1-scores of 78.59% and 59.40% for classifying constructive and toxic comments, respectively. Besides, we implement various baseline models as traditional Machine Learning and Deep Neural Network-Based models to evaluate the dataset. With the results, we can solve several tasks on the online discussions and develop the framework for identifying constructiveness and toxicity of Vietnamese social media comments automatically.
Adversarial attacks are label-preserving modifications to inputs of machine learning classifiers designed to fool machines but not humans. Natural Language Processing (NLP) has mostly focused on high-level attack scenarios such as paraphrasing input texts. We argue that these are less realistic in typical application scenarios such as in social media, and instead focus on low-level attacks on the character-level. Guided by human cognitive abilities and human robustness, we propose the first large-scale catalogue and benchmark of low-level adversarial attacks, which we dub Z\'eroe, encompassing nine different attack modes including visual and phonetic adversaries. We show that RoBERTa, NLP's current workhorse, fails on our attacks. Our dataset provides a benchmark for testing robustness of future more human-like NLP models.
Recent years have witnessed the emergence of a new paradigm of building natural language processing (NLP) systems: general-purpose, pre-trained language models (LMs) are composed with simple downstream models and fine-tuned for a variety of NLP tasks. This paradigm shift significantly simplifies the system development cycles. However, as many LMs are provided by untrusted third parties, their lack of standardization or regulation entails profound security implications, which are largely unexplored. To bridge this gap, this work studies the security threats posed by malicious LMs to NLP systems. Specifically, we present TROJAN-LM, a new class of trojaning attacks in which maliciously crafted LMs trigger host NLP systems to malfunction in a highly predictable manner. By empirically studying three state-of-the-art LMs (BERT, GPT-2, XLNet) in a range of security-critical NLP tasks (toxic comment detection, question answering, text completion) as well as user studies on crowdsourcing platforms, we demonstrate that TROJAN-LM possesses the following properties: (i) flexibility - the adversary is able to flexibly dene logical combinations (e.g., 'and', 'or', 'xor') of arbitrary words as triggers, (ii) efficacy - the host systems misbehave as desired by the adversary with high probability when trigger-embedded inputs are present, (iii) specificity - the trojan LMs function indistinguishably from their benign counterparts on clean inputs, and (iv) fluency - the trigger-embedded inputs appear as fluent natural language and highly relevant to their surrounding contexts. We provide analytical justification for the practicality of TROJAN-LM, and further discuss potential countermeasures and their challenges, which lead to several promising research directions.
This paper tackles one of the greatest limitations in Machine Learning: Data Scarcity. Specifically, we explore whether high accuracy classifiers can be built from small datasets, utilizing a combination of data augmentation techniques and machine learning algorithms. In this paper, we experiment with Easy Data Augmentation (EDA) and Backtranslation, as well as with three popular learning algorithms, Logistic Regression, Support Vector Machine (SVM), and Bidirectional Long Short-Term Memory Network (Bi-LSTM). For our experimentation, we utilize the Wikipedia Toxic Comments dataset so that in the process of exploring the benefits of data augmentation, we can develop a model to detect and classify toxic speech in comments to help fight back against cyberbullying and online harassment. Ultimately, we found that data augmentation techniques can be used to significantly boost the performance of classifiers and are an excellent strategy to combat lack of data in NLP problems.
Toxic comment classification has become an active research field with many recently proposed approaches. However, while these approaches address some of the task's challenges others still remain unsolved and directions for further research are needed. To this end, we compare different deep learning and shallow approaches on a new, large comment dataset and propose an ensemble that outperforms all individual models. Further, we validate our findings on a second dataset. The results of the ensemble enable us to perform an extensive error analysis, which reveals open challenges for state-of-the-art methods and directions towards pending future research. These challenges include missing paradigmatic context and inconsistent dataset labels.