Research shows that natural language processing models are generally considered to be vulnerable to adversarial attacks; but recent work has drawn attention to the issue of validating these adversarial inputs against certain criteria (e.g., the preservation of semantics and grammaticality). Enforcing constraints to uphold such criteria may render attacks unsuccessful, raising the question of whether valid attacks are actually feasible. In this work, we investigate this through the lens of human language ability. We report on crowdsourcing studies in which we task humans with iteratively modifying words in an input text, while receiving immediate model feedback, with the aim of causing a sentiment classification model to misclassify the example. Our findings suggest that humans are capable of generating a substantial amount of adversarial examples using semantics-preserving word substitutions. We analyze how human-generated adversarial examples compare to the recently proposed TextFooler, Genetic, BAE and SememePSO attack algorithms on the dimensions naturalness, preservation of sentiment, grammaticality and substitution rate. Our findings suggest that human-generated adversarial examples are not more able than the best algorithms to generate natural-reading, sentiment-preserving examples, though they do so by being much more computationally efficient.
For many business applications, we often seek to analyze sentiments associated with any arbitrary aspects of commercial products, despite having a very limited amount of labels or even without any labels at all. However, existing aspect target sentiment classification (ATSC) models are not trainable if annotated datasets are not available. Even with labeled data, they fall short of reaching satisfactory performance. To address this, we propose simple approaches that better solve ATSC with natural language prompts, enabling the task under zero-shot cases and enhancing supervised settings, especially for few-shot cases. Under the few-shot setting for SemEval 2014 Task 4 laptop domain, our method of reformulating ATSC as an NLI task outperforms supervised SOTA approaches by up to 24.13 accuracy points and 33.14 macro F1 points. Moreover, we demonstrate that our prompts could handle implicitly stated aspects as well: our models reach about 77% accuracy on detecting sentiments for aspect categories (e.g., food), which do not necessarily appear within the text, even though we trained the models only with explicitly mentioned aspect terms (e.g., fajitas) from just 16 reviews - while the accuracy of the no-prompt baseline is only around 65%.
Aspect-level sentiment classification (ALSC) aims at identifying the sentiment polarity of a specified aspect in a sentence. ALSC is a practical setting in aspect-based sentiment analysis due to no opinion term labeling needed, but it fails to interpret why a sentiment polarity is derived for the aspect. To address this problem, recent works fine-tune pre-trained Transformer encoders for ALSC to extract an aspect-centric dependency tree that can locate the opinion words. However, the induced opinion words only provide an intuitive cue far below human-level interpretability. Besides, the pre-trained encoder tends to internalize an aspect's intrinsic sentiment, causing sentiment bias and thus affecting model performance. In this paper, we propose a span-based anti-bias aspect representation learning framework. It first eliminates the sentiment bias in the aspect embedding by adversarial learning against aspects' prior sentiment. Then, it aligns the distilled opinion candidates with the aspect by span-based dependency modeling to highlight the interpretable opinion terms. Our method achieves new state-of-the-art performance on five benchmarks, with the capability of unsupervised opinion extraction.
The dominating NLP paradigm of training a strong neural predictor to perform one task on a specific dataset has led to state-of-the-art performance in a variety of applications (eg. sentiment classification, span-prediction based question answering or machine translation). However, it builds upon the assumption that the data distribution is stationary, ie. that the data is sampled from a fixed distribution both at training and test time. This way of training is inconsistent with how we as humans are able to learn from and operate within a constantly changing stream of information. Moreover, it is ill-adapted to real-world use cases where the data distribution is expected to shift over the course of a model's lifetime. The first goal of this thesis is to characterize the different forms this shift can take in the context of natural language processing, and propose benchmarks and evaluation metrics to measure its effect on current deep learning architectures. We then proceed to take steps to mitigate the effect of distributional shift on NLP models. To this end, we develop methods based on parametric reformulations of the distributionally robust optimization framework. Empirically, we demonstrate that these approaches yield more robust models as demonstrated on a selection of realistic problems. In the third and final part of this thesis, we explore ways of efficiently adapting existing models to new domains or tasks. Our contribution to this topic takes inspiration from information geometry to derive a new gradient update rule which alleviate catastrophic forgetting issues during adaptation.
As a prominent attribution-based explanation algorithm, Integrated Gradients (IG) is widely adopted due to its desirable explanation axioms and the ease of gradient computation. It measures feature importance by averaging the model's output gradient interpolated along a straight-line path in the input data space. However, such straight-line interpolated points are not representative of text data due to the inherent discreteness of the word embedding space. This questions the faithfulness of the gradients computed at the interpolated points and consequently, the quality of the generated explanations. Here we propose Discretized Integrated Gradients (DIG), which allows effective attribution along non-linear interpolation paths. We develop two interpolation strategies for the discrete word embedding space that generates interpolation points that lie close to actual words in the embedding space, yielding more faithful gradient computation. We demonstrate the effectiveness of DIG over IG through experimental and human evaluations on multiple sentiment classification datasets. We provide the source code of DIG to encourage reproducible research.
Transliteration is very common on social media, but transliterated text is not adequately handled by modern neural models for various NLP tasks. In this work, we combine data augmentation approaches with a Teacher-Student training scheme to address this issue in a cross-lingual transfer setting for fine-tuning state-of-the-art pre-trained multilingual language models such as mBERT and XLM-R. We evaluate our method on transliterated Hindi and Malayalam, also introducing new datasets for benchmarking on real-world scenarios: one on sentiment classification in transliterated Malayalam, and another on crisis tweet classification in transliterated Hindi and Malayalam (related to the 2013 North India and 2018 Kerala floods). Our method yielded an average improvement of +5.6% on mBERT and +4.7% on XLM-R in F1 scores over their strong baselines.
While the whole world is still struggling with the COVID-19 pandemic, online learning and home office become more common. Many schools transfer their courses teaching to the online classroom. Therefore, it is significant to mine the students' feedback and opinions from their reviews towards studies so that both schools and teachers can know where they need to improve. This paper trains machine learning and deep learning models using both balanced and imbalanced datasets for sentiment classification. Two SOTA category-aware text generation GAN models: CatGAN and SentiGAN, are utilized to synthesize text used to balance the highly imbalanced dataset. Results on three datasets with different imbalance degree from distinct domains show that when using generated text to balance the dataset, the F1-score of machine learning and deep learning model on sentiment classification increases 2.79% ~ 9.28%. Also, the results indicate that the average growth degree for CR100k is higher than CR23k, the average growth degree for deep learning is more increased than machine learning algorithms, and the average growth degree for more complex deep learning models is more increased than simpler deep learning models in experiments.
Aspect level sentiment classification (ALSC) is a difficult problem with state-of-the-art models showing less than 80% macro-F1 score on benchmark datasets. Existing models do not incorporate information on aspect-aspect relations in knowledge graphs (KGs), e.g. DBpedia. Two main challenges stem from inaccurate disambiguation of aspects to KG entities, and the inability to learn aspect representations from the large KGs in joint training with ALSC models. We propose AR-BERT, a novel two-level global-local entity embedding scheme that allows efficient joint training of KG-based aspect embeddings and ALSC models. A novel incorrect disambiguation detection technique addresses the problem of inaccuracy in aspect disambiguation. We also introduce the problem of determining mode significance in multi-modal explanation generation, and propose a two step solution. The proposed methods show a consistent improvement of 2.5 - 4.1 percentage points, over the recent BERT-based baselines on benchmark datasets. The code is available at https://github.com/mainuliitkgp/AR-BERT.git.
As an effective approach to tune pre-trained language models (PLMs) for specific tasks, prompt-learning has recently attracted much attention from researchers. By using \textit{cloze}-style language prompts to stimulate the versatile knowledge of PLMs, prompt-learning can achieve promising results on a series of NLP tasks, such as natural language inference, sentiment classification, and knowledge probing. In this work, we investigate the application of prompt-learning on fine-grained entity typing in fully supervised, few-shot and zero-shot scenarios. We first develop a simple and effective prompt-learning pipeline by constructing entity-oriented verbalizers and templates and conducting masked language modeling. Further, to tackle the zero-shot regime, we propose a self-supervised strategy that carries out distribution-level optimization in prompt-learning to automatically summarize the information of entity types. Extensive experiments on three fine-grained entity typing benchmarks (with up to 86 classes) under fully supervised, few-shot and zero-shot settings show that prompt-learning methods significantly outperform fine-tuning baselines, especially when the training data is insufficient.
Activation functions play a pivotal role in determining the training dynamics and neural network performance. The widely adopted activation function ReLU despite being simple and effective has few disadvantages including the Dying ReLU problem. In order to tackle such problems, we propose a novel activation function called Serf which is self-regularized and nonmonotonic in nature. Like Mish, Serf also belongs to the Swish family of functions. Based on several experiments on computer vision (image classification and object detection) and natural language processing (machine translation, sentiment classification and multimodal entailment) tasks with different state-of-the-art architectures, it is observed that Serf vastly outperforms ReLU (baseline) and other activation functions including both Swish and Mish, with a markedly bigger margin on deeper architectures. Ablation studies further demonstrate that Serf based architectures perform better than those of Swish and Mish in varying scenarios, validating the effectiveness and compatibility of Serf with varying depth, complexity, optimizers, learning rates, batch sizes, initializers and dropout rates. Finally, we investigate the mathematical relation between Swish and Serf, thereby showing the impact of preconditioner function ingrained in the first derivative of Serf which provides a regularization effect making gradients smoother and optimization faster.
Reddit.com is a popular social media platform among young people. Reddit users share their stories to seek support from other users, especially during the Covid-19 pandemic. Messages posted on Reddit and their content have provided researchers with opportunity to analyze public concerns. In this study, we analyzed sentiments of COVID-related messages posted on r/Depression. Our study poses the following questions: a) What are the common topics that the Reddit users discuss? b) Can we use these topics to classify sentiments of the posts? c) What matters concern people more during the pandemic? Key Words: Sentiment Classification, Depression, COVID-19, Reddit, LDA, BERT
Aspect-level sentiment classification (ASC) aims to predict the fine-grained sentiment polarity towards a given aspect mentioned in a review. Despite recent advances in ASC, enabling machines to preciously infer aspect sentiments is still challenging. This paper tackles two challenges in ASC: (1) due to lack of aspect knowledge, aspect representation derived in prior works is inadequate to represent aspect's exact meaning and property information; (2) prior works only capture either local syntactic information or global relational information, thus missing either one of them leads to insufficient syntactic information. To tackle these challenges, we propose a novel ASC model which not only end-to-end embeds and leverages aspect knowledge but also marries the two kinds of syntactic information and lets them compensate for each other. Our model includes three key components: (1) a knowledge-aware gated recurrent memory network recurrently integrates dynamically summarized aspect knowledge; (2) a dual syntax graph network combines both kinds of syntactic information to comprehensively capture sufficient syntactic information; (3) a knowledge integrating gate re-enhances the final representation with further needed aspect knowledge; (4) an aspect-to-context attention mechanism aggregates the aspect-related semantics from all hidden states into the final representation. Experimental results on several benchmark datasets demonstrate the effectiveness of our model, which overpass previous state-of-the-art models by large margins in terms of both Accuracy and Macro-F1.
Multimodal target/aspect sentiment classification combines multimodal sentiment analysis and aspect/target sentiment classification. The goal of the task is to combine vision and language to understand the sentiment towards a target entity in a sentence. Twitter is an ideal setting for the task because it is inherently multimodal, highly emotional, and affects real world events. However, multimodal tweets are short and accompanied by complex, possibly irrelevant images. We introduce a two-stream model that translates images in input space using an object-aware transformer followed by a single-pass non-autoregressive text generation approach. We then leverage the translation to construct an auxiliary sentence that provides multimodal information to a language model. Our approach increases the amount of text available to the language model and distills the object-level information in complex images. We achieve state-of-the-art performance on two multimodal Twitter datasets without modifying the internals of the language model to accept multimodal data, demonstrating the effectiveness of our translation. In addition, we explain a failure mode of a popular approach for aspect sentiment analysis when applied to tweets. Our code is available at \textcolor{blue}{\url{https://github.com/codezakh/exploiting-BERT-thru-translation}}.
Multi-modal affect recognition models leverage complementary information in different modalities to outperform their uni-modal counterparts. However, due to the unavailability of modality-specific sensors or data, multi-modal models may not be always employable. For this reason, we aim to improve the performance of uni-modal affect recognition models by transferring knowledge from a better-performing (or stronger) modality to a weaker modality during training. Our proposed multi-modal training framework for cross-modal knowledge transfer relies on two main steps. First, an encoder-classifier model creates task-specific representations for the stronger modality. Then, cross-modal translation generates multi-modal intermediate representations, which are also aligned in the latent space with the stronger modality representations. To exploit the contextual information in temporal sequential affect data, we use Bi-GRU and transformer encoder. We validate our approach on two multi-modal affect datasets, namely CMU-MOSI for binary sentiment classification and RECOLA for dimensional emotion regression. The results show that the proposed approach consistently improves the uni-modal test-time performance of the weaker modalities.
Aspect-based sentiment analysis(ABSA) is a textual analysis methodology that defines the polarity of opinions on certain aspects related to specific targets. The majority of research on ABSA is in English, with a small amount of work available in Arabic. Most previous Arabic research has relied on deep learning models that depend primarily on context-independent word embeddings (e.g.word2vec), where each word has a fixed representation independent of its context. This article explores the modeling capabilities of contextual embeddings from pre-trained language models, such as BERT, and making use of sentence pair input on Arabic aspect sentiment polarity classification task. In particular, we develop a simple but effective BERT-based neural baseline to handle this task. Our BERT architecture with a simple linear classification layer surpassed the state-of-the-art works, according to the experimental results on three different Arabic datasets. Achieving an accuracy of 89.51% on the Arabic hotel reviews dataset, 73% on the Human annotated book reviews dataset, and 85.73% on the Arabic news dataset.
One crucial aspect of sentiment analysis is negation handling, where the occurrence of negation can flip the sentiment of a sentence and negatively affects the machine learning-based sentiment classification. The role of negation in Arabic sentiment analysis has been explored only to a limited extent, especially for colloquial Arabic. In this paper, the author addresses the negation problem of machine learning-based sentiment classification for a colloquial Arabic language. To this end, we propose a simple rule-based algorithm for handling the problem; the rules were crafted based on observing many cases of negation. Additionally, simple linguistic knowledge and sentiment lexicon are used for this purpose. The author also examines the impact of the proposed algorithm on the performance of different machine learning algorithms. The results given by the proposed algorithm are compared with three baseline models. The experimental results show that there is a positive impact on the classifiers accuracy, precision and recall when the proposed algorithm is used compared to the baselines.
Automatically mining sentiment tendency contained in natural language is a fundamental research to some artificial intelligent applications, where solutions alternate with challenges. Transfer learning and multi-task learning techniques have been leveraged to mitigate the supervision sparsity and collaborate multiple heterogeneous domains correspondingly. Recent years, the sensitive nature of users' private data raises another challenge for sentiment classification, i.e., data privacy protection. In this paper, we resort to federated learning for multiple domain sentiment classification under the constraint that the corpora must be stored on decentralized devices. In view of the heterogeneous semantics across multiple parties and the peculiarities of word embedding, we pertinently provide corresponding solutions. First, we propose a Knowledge Transfer Enhanced Private-Shared (KTEPS) framework for better model aggregation and personalization in federated sentiment classification. Second, we propose KTEPS$^\star$ with the consideration of the rich semantic and huge embedding size properties of word vectors, utilizing Projection-based Dimension Reduction (PDR) methods for privacy protection and efficient transmission simultaneously. We propose two federated sentiment classification scenes based on public benchmarks, and verify the superiorities of our proposed methods with abundant experimental investigations.
In this study, we test transfer learning approach on Russian sentiment benchmark datasets using additional train sample created with distant supervision technique. We compare several variants of combining additional data with benchmark train samples. The best results were achieved using three-step approach of sequential training on general, thematic and original train samples. For most datasets, the results were improved by more than 3% to the current state-of-the-art methods. The BERT-NLI model treating sentiment classification problem as a natural language inference task reached the human level of sentiment analysis on one of the datasets.
In this report, address the problem of sentiment classification on the Twitter dataset. used a number of machine learning and deep learning methods to perform sentiment analysis. In the end, used a majority vote ensemble method with 5 of our best models to achieve the classification accuracy of 83.58% on Kaggle public leaderboard.
Aspect Sentiment Triplet Extraction (ASTE) aims to recognize targets, their sentiment polarities and opinions explaining the sentiment from a sentence. ASTE could be naturally divided into 3 atom subtasks, namely target detection, opinion detection and sentiment classification. We argue that the proper subtask combination, compositional feature extraction for target-opinion pairs, and interaction between subtasks would be the key to success. Prior work, however, may fail on `one-to-many' or `many-to-one' situations or derive non-existent sentiment triplets due to defective subtask formulation, sub-optimal feature representation or the lack of subtask interaction. In this paper, we divide ASTE into target-opinion joint detection and sentiment classification subtasks, which is in line with human cognition, and correspondingly utilize sequence encoder and table encoder to handle them. Table encoder extracts sentiment at token-pair level, so that the compositional feature between targets and opinions can be easily captured. To establish explicit interaction between subtasks, we utilize the table representation to guide the sequence encoding, and inject the sequence features back into the table encoder. Experiments show that our model outperforms state-of-the-art methods on six popular ASTE datasets.