In weakly-supervised text classification, only label names act as sources of supervision. Predominant approaches to weakly-supervised text classification utilize a two-phase framework, where test samples are first assigned pseudo-labels and are then used to train a neural text classifier. In most previous work, the pseudo-labeling step is dependent on obtaining seed words that best capture the relevance of each class label. We present LIME, a framework for weakly-supervised text classification that entirely replaces the brittle seed-word generation process with entailment-based pseudo-classification. We find that combining weakly-supervised classification and textual entailment mitigates shortcomings of both, resulting in a more streamlined and effective classification pipeline. With just an off-the-shelf textual entailment model, LIME outperforms recent baselines in weakly-supervised text classification and achieves state-of-the-art in 4 benchmarks. We open source our code at https://github.com/seongminp/LIME.
In this research, we propose a complete set of approaches for identifying and extracting emotions from Bangla texts. We provide a Bangla emotion classifier for six classes: anger, disgust, fear, joy, sadness, and surprise, from Bangla words using transformer-based models, which exhibit phenomenal results in recent days, especially for high-resource languages. The Unified Bangla Multi-class Emotion Corpus (UBMEC) is used to assess the performance of our models. UBMEC is created by combining two previously released manually labeled datasets of Bangla comments on six emotion classes with fresh manually labeled Bangla comments created by us. The corpus dataset and code we used in this work are publicly available.
Graph-based text representation focuses on how text documents are represented as graphs for exploiting dependency information between tokens and documents within a corpus. Despite the increasing interest in graph representation learning, there is limited research in exploring new ways for graph-based text representation, which is important in downstream natural language processing tasks. In this paper, we first propose a new heterogeneous word-character text graph that combines word and character n-gram nodes together with document nodes, allowing us to better learn dependencies among these entities. Additionally, we propose two new graph-based neural models, WCTextGCN and WCTextGAT, for modeling our proposed text graph. Extensive experiments in text classification and automatic text summarization benchmarks demonstrate that our proposed models consistently outperform competitive baselines and state-of-the-art graph-based models.
Annotating large collections of textual data can be time consuming and expensive. That is why the ability to train models with limited annotation budgets is of great importance. In this context, it has been shown that under tight annotation budgets the choice of data representation is key. The goal of this paper is to better understand why this is so. With this goal in mind, we propose a metric that measures the extent to which a given representation is structurally aligned with a task. We conduct experiments on several text classification datasets testing a variety of models and representations. Using our proposed metric we show that an efficient representation for a task (i.e. one that enables learning from few samples) is a representation that induces a good alignment between latent input structure and class structure.
In this work, we combine the two paradigms: Federated Learning (FL) and Continual Learning (CL) for text classification task in cloud-edge continuum. The objective of Federated Continual Learning (FCL) is to improve deep learning models over life time at each client by (relevant and efficient) knowledge transfer without sharing data. Here, we address challenges in minimizing inter-client interference while knowledge sharing due to heterogeneous tasks across clients in FCL setup. In doing so, we propose a novel framework, Federated Selective Inter-client Transfer (FedSeIT) which selectively combines model parameters of foreign clients. To further maximize knowledge transfer, we assess domain overlap and select informative tasks from the sequence of historical tasks at each foreign client while preserving privacy. Evaluating against the baselines, we show improved performance, a gain of (average) 12.4\% in text classification over a sequence of tasks using five datasets from diverse domains. To the best of our knowledge, this is the first work that applies FCL to NLP.
The large majority of the research performed on stance detection has been focused on developing more or less sophisticated text classification systems, even when many benchmarks are based on social network data such as Twitter. This paper aims to take on the stance detection task by placing the emphasis not so much on the text itself but on the interaction data available on social networks. More specifically, we propose a new method to leverage social information such as friends and retweets by generating relational embeddings, namely, dense vector representations of interaction pairs. Our method can be applied to any language and target without any manual tuning. Our experiments on seven publicly available datasets and four different languages show that combining our relational embeddings with textual methods helps to substantially improve performance, obtaining best results for six out of seven evaluation settings, outperforming strong baselines based on large pre-trained language models.
Deep Learning revolutionizes almost all fields of computer science including data management. However, the demand for high-quality training data is slowing down deep neural nets' wider adoption. To this end, data augmentation (DA), which generates more labeled examples from existing ones, becomes a common technique. Meanwhile, the risk of creating noisy examples and the large space of hyper-parameters make DA less attractive in practice. We introduce Rotom, a multi-purpose data augmentation framework for a range of data management and mining tasks including entity matching, data cleaning, and text classification. Rotom features InvDA, a new DA operator that generates natural yet diverse augmented examples by formulating DA as a seq2seq task. The key technical novelty of Rotom is a meta-learning framework that automatically learns a policy for combining examples from different DA operators, whereby combinatorially reduces the hyper-parameters space. Our experimental results show that Rotom effectively improves a model's performance by combining multiple DA operators, even when applying them individually does not yield performance improvement. With this strength, Rotom outperforms the state-of-the-art entity matching and data cleaning systems in the low-resource settings as well as two recently proposed DA techniques for text classification.
Batch Normalization (BN) is a core and prevalent technique in accelerating the training of deep neural networks and improving the generalization on Computer Vision (CV) tasks. However, it fails to defend its position in Natural Language Processing (NLP), which is dominated by Layer Normalization (LN). In this paper, we are trying to answer why BN usually performs worse than LN in NLP tasks with Transformer models. We find that the inconsistency between training and inference of BN is the leading cause that results in the failure of BN in NLP. We define Training Inference Discrepancy (TID) to quantitatively measure this inconsistency and reveal that TID can indicate BN's performance, supported by extensive experiments, including image classification, neural machine translation, language modeling, sequence labeling, and text classification tasks. We find that BN can obtain much better test performance than LN when TID keeps small through training. To suppress the explosion of TID, we propose Regularized BN (RBN) that adds a simple regularization term to narrow the gap between batch statistics and population statistics of BN. RBN improves the performance of BN consistently and outperforms or is on par with LN on 17 out of 20 settings, involving ten datasets and two common variants of Transformer\footnote{Our code is available at \url{https://github.com/wjxts/RegularizedBN}}.
The conventional success of textual classification relies on annotated data, and the new paradigm of pre-trained language models (PLMs) still requires a few labeled data for downstream tasks. However, in real-world applications, label noise inevitably exists in training data, damaging the effectiveness, robustness, and generalization of the models constructed on such data. Recently, remarkable achievements have been made to mitigate this dilemma in visual data, while only a few explore textual data. To fill this gap, we present SelfMix, a simple yet effective method, to handle label noise in text classification tasks. SelfMix uses the Gaussian Mixture Model to separate samples and leverages semi-supervised learning. Unlike previous works requiring multiple models, our method utilizes the dropout mechanism on a single model to reduce the confirmation bias in self-training and introduces a textual-level mixup training strategy. Experimental results on three text classification benchmarks with different types of text show that the performance of our proposed method outperforms these strong baselines designed for both textual and visual data under different noise ratios and noise types. Our code is available at \url{https://github.com/noise-learning/SelfMix}.
Parameter-efficient tuning aims at updating only a small subset of parameters when adapting a pretrained model to downstream tasks. In this work, we introduce PASTA, in which we only modify the special token representations (e.g., [SEP] and [CLS] in BERT) before the self-attention module at each layer in Transformer-based models. PASTA achieves comparable performance to full finetuning in natural language understanding tasks including text classification and NER with up to only 0.029% of total parameters trained. Our work not only provides a simple yet effective way of parameter-efficient tuning, which has a wide range of practical applications when deploying finetuned models for multiple tasks, but also demonstrates the pivotal role of special tokens in pretrained language models
Medical text learning has recently emerged as a promising area to improve healthcare due to the wide adoption of electronic health record (EHR) systems. The complexity of the medical text such as diverse length, mixed text types, and full of medical jargon, poses a great challenge for developing effective deep learning models. BERT has presented state-of-the-art results in many NLP tasks, such as text classification and question answering. However, the standalone BERT model cannot deal with the complexity of the medical text, especially the lengthy clinical notes. Herein, we develop a new model called KG-MTT-BERT (Knowledge Graph Enhanced Multi-Type Text BERT) by extending the BERT model for long and multi-type text with the integration of the medical knowledge graph. Our model can outperform all baselines and other state-of-the-art models in diagnosis-related group (DRG) classification, which requires comprehensive medical text for accurate classification. We also demonstrated that our model can effectively handle multi-type text and the integration of medical knowledge graph can significantly improve the performance.
How can we protect the intellectual property of trained NLP models? Modern NLP models are prone to stealing by querying and distilling from their publicly exposed APIs. However, existing protection methods such as watermarking only work for images but are not applicable to text. We propose Distillation-Resistant Watermarking (DRW), a novel technique to protect NLP models from being stolen via distillation. DRW protects a model by injecting watermarks into the victim's prediction probability corresponding to a secret key and is able to detect such a key by probing a suspect model. We prove that a protected model still retains the original accuracy within a certain bound. We evaluate DRW on a diverse set of NLP tasks including text classification, part-of-speech tagging, and named entity recognition. Experiments show that DRW protects the original model and detects stealing suspects at 100% mean average precision for all four tasks while the prior method fails on two.
Pre-training a large transformer model on a massive amount of unlabeled data and fine-tuning it on labeled datasets for diverse downstream tasks has proven to be a successful strategy, for a variety of vision and natural language processing tasks. However, direct fine-tuning of the pre-trained model may be suboptimal if there exist large discrepancies across data domains for pre-training and fine-tuning. To tackle this issue, several previous studies have proposed further pre-training strategies, where we continue to pre-train the model on the target unlabeled dataset before fine-tuning. However, all of them solely focus on language models and we empirically find that a Vision Transformer is vulnerable to overfitting as we continue to pretrain the model on target unlabeled data. In order to tackle this limitation, we propose self-distillation as a regularization for a further pre-training stage. Specifically, we first further pre-train the initial pre-trained model on the target unlabeled data and then consider it as a teacher for self-distillation. Then we take the same initial pre-trained model as a student and enforce its hidden representations to be close to those of the teacher while optimizing the student with a masked auto-encoding objective. We empirically validate the efficacy of self-distillation on a variety of benchmark datasets for image and text classification tasks. Experimentally, we show that our proposed method outperforms all the relevant baselines. Theoretically, we analyze the proposed method with a simplified model to understand how self-distillation for further pre-training can potentially help improve the performance of the downstream tasks.
To obtain a large amount of training labels inexpensively, researchers have recently adopted the weak supervision (WS) paradigm, which leverages labeling rules to synthesize training labels rather than using individual annotations to achieve competitive results for natural language processing (NLP) tasks. However, data imbalance is often overlooked in applying the WS paradigm, despite being a common issue in a variety of NLP tasks. To address this challenge, we propose Adaptive Ranking-based Sample Selection (ARS2), a model-agnostic framework to alleviate the data imbalance issue in the WS paradigm. Specifically, it calculates a probabilistic margin score based on the output of the current model to measure and rank the cleanliness of each data point. Then, the ranked data are sampled based on both class-wise and rule-aware ranking. In particular, the two sample strategies corresponds to our motivations: (1) to train the model with balanced data batches to reduce the data imbalance issue and (2) to exploit the expertise of each labeling rule for collecting clean samples. Experiments on four text classification datasets with four different imbalance ratios show that ARS2 outperformed the state-of-the-art imbalanced learning and WS methods, leading to a 2%-57.8% improvement on their F1-score.
Text augmentation is an effective technique for addressing the problem of insufficient data in natural language processing. However, existing text augmentation methods tend to focus on few-shot scenarios and usually perform poorly on large public datasets. Our research indicates that existing augmentation methods often generate instances with shifted feature spaces, which leads to a drop in performance on the augmented data (for example, EDA generally loses $\approx 2\%$ in aspect-based sentiment classification). To address this problem, we propose a hybrid instance-filtering framework (BootAug) based on pre-trained language models that can maintain a similar feature space with natural datasets. BootAug is transferable to existing text augmentation methods (such as synonym substitution and back translation) and significantly improves the augmentation performance by $\approx 2-3\%$ in classification accuracy. Our experimental results on three classification tasks and nine public datasets show that BootAug addresses the performance drop problem and outperforms state-of-the-art text augmentation methods. Additionally, we release the code to help improve existing augmentation methods on large datasets.
Embeddings, which compress information in raw text into semantics-preserving low-dimensional vectors, have been widely adopted for their efficacy. However, recent research has shown that embeddings can potentially leak private information about sensitive attributes of the text, and in some cases, can be inverted to recover the original input text. To address these growing privacy challenges, we propose a privatization mechanism for embeddings based on homomorphic encryption, to prevent potential leakage of any piece of information in the process of text classification. In particular, our method performs text classification on the encryption of embeddings from state-of-the-art models like BERT, supported by an efficient GPU implementation of CKKS encryption scheme. We show that our method offers encrypted protection of BERT embeddings, while largely preserving their utility on downstream text classification tasks.
Abbreviations are unavoidable yet critical parts of the medical text. Using abbreviations, especially in clinical patient notes, can save time and space, protect sensitive information, and help avoid repetitions. However, most abbreviations might have multiple senses, and the lack of a standardized mapping system makes disambiguating abbreviations a difficult and time-consuming task. The main objective of this study is to examine the feasibility of token classification methods for medical abbreviation disambiguation. Specifically, we explore the capability of token classification methods to deal with multiple unique abbreviations in a single text. We use two public datasets to compare and contrast the performance of several transformer models pre-trained on different scientific and medical corpora. Our proposed token classification approach outperforms the more commonly used text classification models for the abbreviation disambiguation task. In particular, the SciBERT model shows a strong performance for both token and text classification tasks over the two considered datasets. Furthermore, we find that abbreviation disambiguation performance for the text classification models becomes comparable to that of token classification only when postprocessing is applied to their predictions, which involves filtering possible labels for an abbreviation based on the training data.
Transfer learning from large language models (LLMs) has emerged as a powerful technique to enable knowledge-based fine-tuning for a number of tasks, adaptation of models for different domains and even languages. However, it remains an open question, if and when transfer learning will work, i.e. leading to positive or negative transfer. In this paper, we analyze the knowledge transfer across three natural language processing (NLP) tasks - text classification, sentimental analysis, and sentence similarity, using three LLMs - BERT, RoBERTa, and XLNet - and analyzing their performance, by fine-tuning on target datasets for domain and cross-lingual adaptation tasks, with and without an intermediate task training on a larger dataset. Our experiments showed that fine-tuning without an intermediate task training can lead to a better performance for most tasks, while more generalized tasks might necessitate a preceding intermediate task training step. We hope that this work will act as a guide on transfer learning to NLP practitioners.
Standard approaches to hate speech detection rely on sufficient available hate speech annotations. Extending previous work that repurposes natural language inference (NLI) models for zero-shot text classification, we propose a simple approach that combines multiple hypotheses to improve English NLI-based zero-shot hate speech detection. We first conduct an error analysis for vanilla NLI-based zero-shot hate speech detection and then develop four strategies based on this analysis. The strategies use multiple hypotheses to predict various aspects of an input text and combine these predictions into a final verdict. We find that the zero-shot baseline used for the initial error analysis already outperforms commercial systems and fine-tuned BERT-based hate speech detection models on HateCheck. The combination of the proposed strategies further increases the zero-shot accuracy of 79.4% on HateCheck by 7.9 percentage points (pp), and the accuracy of 69.6% on ETHOS by 10.0pp.
Given the wide and ever growing range of different efficient Transformer attention mechanisms, it is important to identify which attention is most effective when given a task. In this work, we are also interested in combining different attention types to build heterogeneous Transformers. We first propose a DARTS-like Neural Architecture Search (NAS) method to find the best attention for a given task, in this setup, all heads use the same attention (homogeneous models). Our results suggest that NAS is highly effective on this task, and it identifies the best attention mechanisms for IMDb byte level text classification and Listops. We then extend our framework to search for and build Transformers with multiple different attention types, and call them heterogeneous Transformers. We show that whilst these heterogeneous Transformers are better than the average homogeneous models, they cannot outperform the best. We explore the reasons why heterogeneous attention makes sense, and why it ultimately fails.