The current success of deep neural networks (DNNs) in an increasingly broad range of tasks involving artificial intelligence strongly depends on the quality and quantity of labeled training data. In general, the scarcity of labeled data, which is often observed in many natural language processing tasks, is one of the most important issues to be addressed. Semi-supervised learning (SSL) is a promising approach to overcoming this issue by incorporating a large amount of unlabeled data. In this paper, we propose a novel scalable method of SSL for text classification tasks. The unique property of our method, Mixture of Expert/Imitator Networks, is that imitator networks learn to "imitate" the estimated label distribution of the expert network over the unlabeled data, which potentially contributes a set of features for the classification. Our experiments demonstrate that the proposed method consistently improves the performance of several types of baseline DNNs. We also demonstrate that our method has the more data, better performance property with promising scalability to the amount of unlabeled data.
We introduce a new language representation model called BERT, which stands for Bidirectional Encoder Representations from Transformers. Unlike recent language representation models, BERT is designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers. As a result, the pre-trained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks, such as question answering and language inference, without substantial task-specific architecture modifications. BERT is conceptually simple and empirically powerful. It obtains new state-of-the-art results on eleven natural language processing tasks, including pushing the GLUE score to 80.5% (7.7% point absolute improvement), MultiNLI accuracy to 86.7% (4.6% absolute improvement), SQuAD v1.1 question answering Test F1 to 93.2 (1.5 point absolute improvement) and SQuAD v2.0 Test F1 to 83.1 (5.1 point absolute improvement).
Text analytics based on supervised machine learning classifiers has shown great promise in a multitude of domains, but has yet to be applied to Seismology. We test various standard models (Naive Bayes, k-Nearest Neighbors, Support Vector Machines, and Random Forests) on a seismological corpus of 100 articles related to the topic of precursory accelerating seismicity, spanning from 1988 to 2010. This corpus was labelled in Mignan (2011) with the precursor whether explained by critical processes (i.e., cascade triggering) or by other processes (such as signature of main fault loading). We investigate rather the classification process can be automatized to help analyze larger corpora in order to better understand trends in earthquake predictability research. We find that the Naive Bayes model performs best, in agreement with the machine learning literature for the case of small datasets, with cross-validation accuracies of 86% for binary classification. For a refined multiclass classification ('non-critical process' < 'agnostic' < 'critical process assumed' < 'critical process demonstrated'), we obtain up to 78% accuracy. Prediction on a dozen of articles published since 2011 shows however a weak generalization with a F1-score of 60%, only slightly better than a random classifier, which can be explained by a change of authorship and use of different terminologies. Yet, the model shows F1-scores greater than 80% for the two multiclass extremes ('non-critical process' versus 'critical process demonstrated') while it falls to random classifier results (around 25%) for papers labelled 'agnostic' or 'critical process assumed'. Those results are encouraging in view of the small size of the corpus and of the high degree of abstraction of the labelling. Domain knowledge engineering remains essential but can be made transparent by an investigation of Naive Bayes keyword posterior probabilities.
For analysing and/or understanding languages having no word boundaries based on morphological analysis such as Japanese, Chinese, and Thai, it is desirable to perform appropriate word segmentation before word embeddings. But it is inherently difficult in these languages. In recent years, various language models based on deep learning have made remarkable progress, and some of these methodologies utilizing character-level features have successfully avoided such a difficult problem. However, when a model is fed character-level features of the above languages, it often causes overfitting due to a large number of character types. In this paper, we propose a CE-CLCNN, character-level convolutional neural networks using a character encoder to tackle these problems. The proposed CE-CLCNN is an end-to-end learning model and has an image-based character encoder, i.e. the CE-CLCNN handles each character in the target document as an image. Through various experiments, we found and confirmed that our CE-CLCNN captured closely embedded features for visually and semantically similar characters and achieves state-of-the-art results on several open document classification tasks. In this paper we report the performance of our CE-CLCNN with the Wikipedia title estimation task and analyse the internal behaviour.
Modern NLP applications have enjoyed a great boost utilizing neural networks models. Such deep neural models, however, are not applicable to most human languages due to the lack of annotated training data for various NLP tasks. Cross-lingual transfer learning (CLTL) is a viable method for building NLP models for a low-resource target language by leveraging labeled data from other (source) languages. In this work, we focus on the multilingual transfer setting where training data in multiple source languages is leveraged to further boost target language performance. Unlike most existing methods that rely only on language-invariant features for CLTL, our approach coherently utilizes both language-invariant and language-specific features at instance level. Our model leverages adversarial networks to learn language-invariant features, and mixture-of-experts models to dynamically exploit the similarity between the target language and each individual source language. This enables our model to learn effectively what to share between various languages in the multilingual setup. Moreover, when coupled with unsupervised multilingual embeddings, our model can operate in a zero-resource setting where neither target language training data nor cross-lingual resources are available. Our model achieves significant performance gains over prior art, as shown in an extensive set of experiments over multiple text classification and sequence tagging tasks including a large-scale industry dataset.
In this paper, we study counterfactual fairness in text classification, which asks the question: How would the prediction change if the sensitive attribute referenced in the example were different? Toxicity classifiers demonstrate a counterfactual fairness issue by predicting that "Some people are gay" is toxic while "Some people are straight" is nontoxic. We offer a metric, counterfactual token fairness (CTF), for measuring this particular form of fairness in text classifiers, and describe its relationship with group fairness. Further, we offer three approaches, blindness, counterfactual augmentation, and counterfactual logit pairing (CLP), for optimizing counterfactual token fairness during training, bridging the robustness and fairness literature. Empirically, we find that blindness and CLP address counterfactual token fairness. The methods do not harm classifier performance, and have varying tradeoffs with group fairness. These approaches, both for measurement and optimization, provide a new path forward for addressing fairness concerns in text classification.
We present an analysis into the inner workings of Convolutional Neural Networks (CNNs) for processing text. CNNs used for computer vision can be interpreted by projecting filters into image space, but for discrete sequence inputs CNNs remain a mystery. We aim to understand the method by which the networks process and classify text. We examine common hypotheses to this problem: that filters, accompanied by global max-pooling, serve as ngram detectors. We show that filters may capture several different semantic classes of ngrams by using different activation patterns, and that global max-pooling induces behavior which separates important ngrams from the rest. Finally, we show practical use cases derived from our findings in the form of model interpretability (explaining a trained model by deriving a concrete identity for each filter, bridging the gap between visualization tools in vision tasks and NLP) and prediction interpretability (explaining predictions). Code implementation is available online at github.com/sayaendo/interpreting-cnn-for-text.
Continuous word representation (aka word embedding) is a basic building block in many neural network-based models used in natural language processing tasks. Although it is widely accepted that words with similar semantics should be close to each other in the embedding space, we find that word embeddings learned in several tasks are biased towards word frequency: the embeddings of high-frequency and low-frequency words lie in different subregions of the embedding space, and the embedding of a rare word and a popular word can be far from each other even if they are semantically similar. This makes learned word embeddings ineffective, especially for rare words, and consequently limits the performance of these neural network models. In this paper, we develop a neat, simple yet effective way to learn \emph{FRequency-AGnostic word Embedding} (FRAGE) using adversarial training. We conducted comprehensive studies on ten datasets across four natural language processing tasks, including word similarity, language modeling, machine translation and text classification. Results show that with FRAGE, we achieve higher performance than the baselines in all tasks.
Automatic judgment prediction aims to predict the judicial results based on case materials. It has been studied for several decades mainly by lawyers and judges, considered as a novel and prospective application of artificial intelligence techniques in the legal field. Most existing methods follow the text classification framework, which fails to model the complex interactions among complementary case materials. To address this issue, we formalize the task as Legal Reading Comprehension according to the legal scenario. Following the working protocol of human judges, LRC predicts the final judgment results based on three types of information, including fact description, plaintiffs' pleas, and law articles. Moreover, we propose a novel LRC model, AutoJudge, which captures the complex semantic interactions among facts, pleas, and laws. In experiments, we construct a real-world civil case dataset for LRC. Experimental results on this dataset demonstrate that our model achieves significant improvement over state-of-the-art models. We will publish all source codes and datasets of this work on \urlgithub.com for further research.
Cross-domain text classification aims at building a classifier for a target domain which leverages data from both source and target domain. One promising idea is to minimize the feature distribution differences of the two domains. Most existing studies explicitly minimize such differences by an exact alignment mechanism (aligning features by one-to-one feature alignment, projection matrix etc.). Such exact alignment, however, will restrict models' learning ability and will further impair models' performance on classification tasks when the semantic distributions of different domains are very different. To address this problem, we propose a novel group alignment which aligns the semantics at group level. In addition, to help the model learn better semantic groups and semantics within these groups, we also propose a partial supervision for model's learning in source domain. To this end, we embed the group alignment and a partial supervision into a cross-domain topic model, and propose a Cross-Domain Labeled LDA (CDL-LDA). On the standard 20Newsgroup and Reuters dataset, extensive quantitative (classification, perplexity etc.) and qualitative (topic detection) experiments are conducted to show the effectiveness of the proposed group alignment and partial supervision.
Text classification is an important and classical problem in natural language processing. There have been a number of studies that applied convolutional neural networks (convolution on regular grid, e.g., sequence) to classification. However, only a limited number of studies have explored the more flexible graph convolutional neural networks (convolution on non-grid, e.g., arbitrary graph) for the task. In this work, we propose to use graph convolutional networks for text classification. We build a single text graph for a corpus based on word co-occurrence and document word relations, then learn a Text Graph Convolutional Network (Text GCN) for the corpus. Our Text GCN is initialized with one-hot representation for word and document, it then jointly learns the embeddings for both words and documents, as supervised by the known class labels for documents. Our experimental results on multiple benchmark datasets demonstrate that a vanilla Text GCN without any external word embeddings or knowledge outperforms state-of-the-art methods for text classification. On the other hand, Text GCN also learns predictive word and document embeddings. In addition, experimental results show that the improvement of Text GCN over state-of-the-art comparison methods become more prominent as we lower the percentage of training data, suggesting the robustness of Text GCN to less training data in text classification.
Many classification models work poorly on short texts due to data sparsity. To address this issue, we propose topic memory networks for short text classification with a novel topic memory mechanism to encode latent topic representations indicative of class labels. Different from most prior work that focuses on extending features with external knowledge or pre-trained topics, our model jointly explores topic inference and text classification with memory networks in an end-to-end manner. Experimental results on four benchmark datasets show that our model outperforms state-of-the-art models on short text classification, meanwhile generates coherent topics.
Cross-lingual transfer of word embeddings aims to establish the semantic mappings among words in different languages by learning the transformation functions over the corresponding word embedding spaces. Successfully solving this problem would benefit many downstream tasks such as to translate text classification models from resource-rich languages (e.g. English) to low-resource languages. Supervised methods for this problem rely on the availability of cross-lingual supervision, either using parallel corpora or bilingual lexicons as the labeled data for training, which may not be available for many low resource languages. This paper proposes an unsupervised learning approach that does not require any cross-lingual labeled data. Given two monolingual word embedding spaces for any language pair, our algorithm optimizes the transformation functions in both directions simultaneously based on distributional matching as well as minimizing the back-translation losses. We use a neural network implementation to calculate the Sinkhorn distance, a well-defined distributional similarity measure, and optimize our objective through back-propagation. Our evaluation on benchmark datasets for bilingual lexicon induction and cross-lingual word similarity prediction shows stronger or competitive performance of the proposed method compared to other state-of-the-art supervised and unsupervised baseline methods over many language pairs.
Industry datasets used for text classification are rarely created for that purpose. In most cases, the data and target predictions are a by-product of accumulated historical data, typically fraught with noise, present in both the text-based document, as well as in the targeted labels. In this work, we address the question of how well performance metrics computed on noisy, historical data reflect the performance on the intended future machine learning model input. The results demonstrate the utility of dirty training datasets used to build prediction models for cleaner (and different) prediction inputs.
In this report, we present a study of eight corpora of online hate speech, by demonstrating the NLP techniques that we used to collect and analyze the jihadist, extremist, racist, and sexist content. Analysis of the multilingual corpora shows that the different contexts share certain characteristics in their hateful rhetoric. To expose the main features, we have focused on text classification, text profiling, keyword and collocation extraction, along with manual annotation and qualitative study.
Local model interpretation methods explain individual predictions by assigning an importance value to each input feature. This value is often determined by measuring the change in confidence when a feature is removed. However, the confidence of neural networks is not a robust measure of model uncertainty. This issue makes reliably judging the importance of the input features difficult. We address this by changing the test-time behavior of neural networks using Deep k-Nearest Neighbors. Without harming text classification accuracy, this algorithm provides a more robust uncertainty metric which we use to generate feature importance values. The resulting interpretations better align with human perception than baseline methods. Finally, we use our interpretation method to analyze model predictions on dataset annotation artifacts.
Multi-label text classification (MLTC) aims to assign multiple labels to each sample in the dataset. The labels usually have internal correlations. However, traditional methods tend to ignore the correlations between labels. In order to capture the correlations between labels, the sequence-to-sequence (Seq2Seq) model views the MLTC task as a sequence generation problem, which achieves excellent performance on this task. However, the Seq2Seq model is not suitable for the MLTC task in essence. The reason is that it requires humans to predefine the order of the output labels, while some of the output labels in the MLTC task are essentially an unordered set rather than an ordered sequence. This conflicts with the strict requirement of the Seq2Seq model for the label order. In this paper, we propose a novel sequence-to-set framework utilizing deep reinforcement learning, which not only captures the correlations between labels, but also reduces the dependence on the label order. Extensive experimental results show that our proposed method outperforms the competitive baselines by a large margin.
Adversarial Reprogramming has demonstrated success in utilizing pre-trained neural network classifiers for alternative classification tasks without modification to the original network. An adversary in such an attack scenario trains an additive contribution to the inputs to repurpose the neural network for the new classification task. While this reprogramming approach works for neural networks with a continuous input space such as that of images, it is not directly applicable to neural networks trained for tasks such as text classification, where the input space is discrete. Repurposing such classification networks would require the attacker to learn an adversarial program that maps inputs from one discrete space to the other. In this work, we introduce a context-based vocabulary remapping model to reprogram neural networks trained on a specific sequence classification task, for a new sequence classification task desired by the adversary. We propose training procedures for this adversarial program in both white-box and black-box settings. We demonstrate the application of our model by adversarially repurposing various text-classification models including LSTM, bi-directional LSTM and CNN for alternate classification tasks.
Efficient distributed numerical word representation models (word embeddings) combined with modern machine learning algorithms have recently yielded considerable improvement on automatic document classification tasks. However, the effectiveness of such techniques has not been assessed for the hierarchical text classification (HTC) yet. This study investigates the application of those models and algorithms on this specific problem by means of experimentation and analysis. We trained classification models with prominent machine learning algorithm implementations---fastText, XGBoost, SVM, and Keras' CNN---and noticeable word embeddings generation methods---GloVe, word2vec, and fastText---with publicly available data and evaluated them with measures specifically appropriate for the hierarchical context. FastText achieved an ${}_{LCA}F_1$ of 0.893 on a single-labeled version of the RCV1 dataset. An analysis indicates that using word embeddings and its flavors is a very promising approach for HTC.
Deep neural networks are gaining increasing popularity for the classic text classification task, due to their strong expressive power and less requirement for feature engineering. Despite such attractiveness, neural text classification models suffer from the lack of training data in many real-world applications. Although many semi-supervised and weakly-supervised text classification models exist, they cannot be easily applied to deep neural models and meanwhile support limited supervision types. In this paper, we propose a weakly-supervised method that addresses the lack of training data in neural text classification. Our method consists of two modules: (1) a pseudo-document generator that leverages seed information to generate pseudo-labeled documents for model pre-training, and (2) a self-training module that bootstraps on real unlabeled data for model refinement. Our method has the flexibility to handle different types of weak supervision and can be easily integrated into existing deep neural models for text classification. We have performed extensive experiments on three real-world datasets from different domains. The results demonstrate that our proposed method achieves inspiring performance without requiring excessive training data and outperforms baseline methods significantly.