Contextualized Word Representations Research Articles

A deep learning method DCWR with HANet for stock market prediction using news articles

Stock movement prediction is a challenging problem to analyze in both academic and financial research areas. The advancement of deep learning (DL) techniques has grasped the attention of researchers to employ them for predicting the stock market’s future trends. Few frameworks can understand the financial terms in literature, and the volatile nature of stock markets further complicates this process. This paper has tried to overcome the existing challenges by introducing a DL-based framework using financial news articles to forecast the stock market. After performing preprocessing step, the deep contextualized word representation (DCWR) approach is applied for feature extraction. In the next step, the independent component analysis (ICA) method is used for feature reduction. Finally, the resultant features train the hierarchical attention networks (HANet) classifier to predict the stock movements. The proposed scheme is evaluated over the 7 years of data from a publicly available dataset gathered from the Reuter’s website and attained an average prediction accuracy of 92.5% which shows our framework’s robustness.

ACNN-TL: attention-based convolutional neural network coupling with transfer learning and contextualized word representation for enhancing the performance of sentiment classification

ACNN-TL: attention-based convolutional neural network coupling with transfer learning and contextualized word representation for enhancing the performance of sentiment classification

Multi-domain Spoken Language Understanding Using Domain- and Task-aware Parameterization

Spoken language understanding (SLU) has been addressed as a supervised learning problem, where a set of training data is available for each domain. However, annotating data for a new domain can be both financially costly and non-scalable. One existing approach solves the problem by conducting multi-domain learning where parameters are shared for joint training across domains, which is domain-agnostic and task-agnostic . In the article, we propose to improve the parameterization of this method by using domain-specific and task-specific model parameters for fine-grained knowledge representation and transfer. Experiments on five domains show that our model is more effective for multi-domain SLU and obtain the best results. In addition, we show its transferability when adapting to a new domain with little data, outperforming the prior best model by 12.4%. Finally, we explore the strong pre-trained model in our framework and find that the contributions from our framework do not fully overlap with contextualized word representations (RoBERTa).

LMMS reloaded: Transformer-based sense embeddings for disambiguation and beyond

Distributional semantics based on neural approaches is a cornerstone of Natural Language Processing, with surprising connections to human meaning representation as well. Recent Transformer-based Language Models have proven capable of producing contextual word representations that reliably convey sense-specific information, simply as a product of self-supervision. Prior work has shown that these contextual representations can be used to accurately represent large sense inventories as sense embeddings, to the extent that a distance-based solution to Word Sense Disambiguation (WSD) tasks outperforms models trained specifically for the task. Still, there remains much to understand on how to use these Neural Language Models (NLMs) to produce sense embeddings that can better harness each NLM's meaning representation abilities. In this work we introduce a more principled approach to leverage information from all layers of NLMs, informed by a probing analysis on 14 NLM variants. We also emphasize the versatility of these sense embeddings in contrast to task-specific models, applying them on several sense-related tasks, besides WSD, while demonstrating improved performance using our proposed approach over prior work focused on sense embeddings. Finally, we discuss unexpected findings regarding layer and model performance variations, and potential applications for downstream tasks.

Investigating Maps of Science Using Contextual Proximity of Citations Based on Deep Contextualized Word Representation

The citation intent extraction and classification has long been studied as it is a good measure of relevancy. Different approaches have classified the citations into different classes; including weak and strong, positive and negative, important and unimportant. Others have gone further from binary classification to multi-classes, including extension, use, background, or comparison. Researchers have utilized various elements of the information, including both meta and contents of the paper. The actual context of any referred article lies within the citation context where a paper is referred. Various attempts have been made to study the citation context to capture the citation intent, but very few have encoded the words to their contextual representations. For automated classification, we need to train deep learning models, which take the citation context as input and provides the reason for citing a paper. Deep neural models work on numeric data, and therefore, we must convert the text information to its numeric representation. Natural languages are much complex than computer languages. Computer languages have a pre-defined fixed syntax where each word has a unique meaning. In contrast, every word in natural language may have a different meaning and may well be understood by understanding the position, previous discussion, and neighboring words. The extra information provides the context of a word within a sentence. We have, therefore, used contextual word representation, which is trained through deep neural networks. Deep models require massive data for generalizing the model, however, the existing state-of-the-art datasets don’t provide much information for the training models to get generalized. Therefore, we have developed our own scholarly dataset, Citation Context Dataset with Intent (C2D-I), an extension of the C2D dataset. We used a transformers based model for capturing the contextual representation of words. Our proposed method outperformed the existing benchmark methods with F1 score of 89%.

Which Apple Keeps Which Doctor Away? Colorful Word Representations With Visual Oracles

Recent pre-trained language models (PrLMs) offer a new performant method of contextualized word representations by leveraging the sequence-level context for modeling. Although the PrLMs generally provide more effective contextualized word representations than non-contextualized models, they are still subject to a sequence of text contexts without diverse hints from multimodality. This paper thus proposes a visual representation method to explicitly enhance conventional word embedding with multiple-aspect senses from visual guidance. In detail, we build a small-scale word-image dictionary from a multimodal seed dataset where each word corresponds to diverse related images. Experiments on 12 natural language understanding and machine translation tasks further verify the effectiveness and the generalization capability of the proposed approach. Analysis shows that our method with visual guidance pays more attention to content words, improves the representation diversity, and is potentially beneficial for enhancing the accuracy of disambiguation.

Visual Comparison of Language Model Adaptation.

Neural language models are widely used; however, their model parameters often need to be adapted to the specific domains and tasks of an application, which is time- and resource-consuming. Thus, adapters have recently been introduced as a lightweight alternative for model adaptation. They consist of a small set of task-specific parameters with a reduced training time and simple parameter composition. The simplicity of adapter training and composition comes along with new challenges, such as maintaining an overview of adapter properties and effectively comparing their produced embedding spaces. To help developers overcome these challenges, we provide a twofold contribution. First, in close collaboration with NLP researchers, we conducted a requirement analysis for an approach supporting adapter evaluation and detected, among others, the need for both intrinsic (i.e., embedding similarity-based) and extrinsic (i.e., prediction-based) explanation methods. Second, motivated by the gathered requirements, we designed a flexible visual analytics workspace that enables the comparison of adapter properties. In this paper, we discuss several design iterations and alternatives for interactive, comparative visual explanation methods. Our comparative visualizations show the differences in the adapted embedding vectors and prediction outcomes for diverse human-interpretable concepts (e.g., person names, human qualities). We evaluate our workspace through case studies and show that, for instance, an adapter trained on the language debiasing task according to context-0 (decontextualized) embeddings introduces a new type of bias where words (even gender-independent words such as countries) become more similar to female- than male pronouns. We demonstrate that these are artifacts of context-0 embeddings, and the adapter effectively eliminates the gender information from the contextualized word representations.

Non-Autoregressive ASR Modeling Using Pre-Trained Language Models for Chinese Speech Recognition

Transformer-based models have led to significant innovation in various classic and practical subjects, including speech processing, natural language processing, and computer vision. On top of the Transformer, attention-based end-to-end automatic speech recognition (ASR) models have become a popular fashion in recent years. Specifically, an emergent research topic is non-autoregressive modeling, which can achieve fast inference speed and obtain competitive performance when compared with conventional autoregressive methods. In addition, in the context of natural language processing, the bidirectional encoder representations from Transformers (BERT) model and its variants have received widespread attention, partially due to their ability to infer contextualized word representations and obtain superior performances of downstream tasks through simple fine-tuning. However, to our knowledge, leveraging the synergistic power of non-autoregressive modeling and pre-trained language model for ASR remains relatively underexplored. In this regard, this study presents a novel pre-trained language model-based non-autoregressive ASR framework. A series of experiments were conducted on two publicly available Chinese datasets, AISHELL-1 and AISHELL-2, to demonstrate competitive or superior results of the proposed ASR models when compared with well-practiced baseline systems. In addition, a set of comparative experiments is likewise carried out with different settings to analyze the performance of the proposed framework.

Analyzing transfer learning impact in biomedical cross-lingual named entity recognition and normalization

BackgroundThe volume of biomedical literature and clinical data is growing at an exponential rate. Therefore, efficient access to data described in unstructured biomedical texts is a crucial task for the biomedical industry and research. Named Entity Recognition (NER) is the first step for information and knowledge acquisition when we deal with unstructured texts. Recent NER approaches use contextualized word representations as input for a downstream classification task. However, distributed word vectors (embeddings) are very limited in Spanish and even more for the biomedical domain.MethodsIn this work, we develop several biomedical Spanish word representations, and we introduce two Deep Learning approaches for pharmaceutical, chemical, and other biomedical entities recognition in Spanish clinical case texts and biomedical texts, one based on a Bi-STM-CRF model and the other on a BERT-based architecture.ResultsSeveral Spanish biomedical embeddigns together with the two deep learning models were evaluated on the PharmaCoNER and CORD-19 datasets. The PharmaCoNER dataset is composed of a set of Spanish clinical cases annotated with drugs, chemical compounds and pharmacological substances; our extended Bi-LSTM-CRF model obtains an F-score of 85.24% on entity identification and classification and the BERT model obtains an F-score of 88.80% . For the entity normalization task, the extended Bi-LSTM-CRF model achieves an F-score of 72.85% and the BERT model achieves 79.97%. The CORD-19 dataset consists of scholarly articles written in English annotated with biomedical concepts such as disorder, species, chemical or drugs, gene and protein, enzyme and anatomy. Bi-LSTM-CRF model and BERT model obtain an F-measure of 78.23% and 78.86% on entity identification and classification, respectively on the CORD-19 dataset.ConclusionThese results prove that deep learning models with in-domain knowledge learned from large-scale datasets highly improve named entity recognition performance. Moreover, contextualized representations help to understand complexities and ambiguity inherent to biomedical texts. Embeddings based on word, concepts, senses, etc. other than those for English are required to improve NER tasks in other languages.

Embeddings in Natural Language Processing: Theory and Advances in Vector Representations of Meaning

Embeddings in Natural Language Processing: Theory and Advances in Vector Representations of Meaning

Anonymization of German financial documents using neural network-based language models with contextual word representations

The automatization and digitalization of business processes have led to an increase in the need for efficient information extraction from business documents. However, financial and legal documents are often not utilized effectively by text processing or machine learning systems, partly due to the presence of sensitive information in these documents, which restrict their usage beyond authorized parties and purposes. To overcome this limitation, we develop an anonymization method for German financial and legal documents using state-of-the-art natural language processing methods based on recurrent neural nets and transformer architectures. We present a web-based application to anonymize financial documents and a large-scale evaluation of different deep learning techniques.

Prediction of user loyalty in mobile applications using deep contextualized word representations

ABSTRACT Customer loyalty is important for many industries, including banking, telecommunications, gaming, and shopping, in terms of sustainability. In mobile applications, it is observed that the demand rises with the usage of mobile devices such as smartphones. Therefore, it is important to predict when players tend to leave an application. Most of the studies so far focus on churn prediction or customer loyalty in mobile applications by analyzing demographic, economic, and behavioral data about customers. In this work, we introduce sentiment analysis-based customer loyalty prediction in mobile applications using word embeddings, deep learning algorithms, and deep contextualized word representations. To our knowledge, this is the first study to evaluate loyalty of customers analyzing sentiments of users from their comments using deep learning, word embedding, and deep contextualized word representation models. For this purpose, CNNs, RNNs, LSTMs, BERT, MBERT, DistilBERT, RoBERT are used for classification purpose. On the other hand, word embedding models such as Word2Vec, GloVe, and FastText are employed for text representation. To demonstrate the impact of proposed model, comprehensive experiments are performed on seven different datasets. The experiment results show sentiment analysis of users in mobile applications can be a powerful indicator in terms of predicting customer loyalty.

Unsupervised cross-lingual model transfer for named entity recognition with contextualized word representations.

Named entity recognition (NER) is one fundamental task in the natural language processing (NLP) community. Supervised neural network models based on contextualized word representations can achieve highly-competitive performance, which requires a large-scale manually-annotated corpus for training. While for the resource-scarce languages, the construction of such as corpus is always expensive and time-consuming. Thus, unsupervised cross-lingual transfer is one good solution to address the problem. In this work, we investigate the unsupervised cross-lingual NER with model transfer based on contextualized word representations, which greatly advances the cross-lingual NER performance. We study several model transfer settings of the unsupervised cross-lingual NER, including (1) different types of the pretrained transformer-based language models as input, (2) the exploration strategies of the multilingual contextualized word representations, and (3) multi-source adaption. In particular, we propose an adapter-based word representation method combining with parameter generation network (PGN) better to capture the relationship between the source and target languages. We conduct experiments on a benchmark ConLL dataset involving four languages to simulate the cross-lingual setting. Results show that we can obtain highly-competitive performance by cross-lingual model transfer. In particular, our proposed adapter-based PGN model can lead to significant improvements for cross-lingual NER.

Temporal Relation Extraction in Clinical Texts

Unstructured data in electronic health records, represented by clinical texts, are a vast source of healthcare information because they describe a patient's journey, including clinical findings, procedures, and information about the continuity of care. The publication of several studies on temporal relation extraction from clinical texts during the last decade and the realization of multiple shared tasks highlight the importance of this research theme. Therefore, we propose a review of temporal relation extraction in clinical texts. We analyzed 105 articles and verified that relations between events and document creation time, a coarse temporality type, were addressed with traditional machine learning–based models with few recent initiatives to push the state-of-the-art with deep learning–based models. For temporal relations between entities (event and temporal expressions) in the document, factors such as dataset imbalance because of candidate pair generation and task complexity directly affect the system's performance. The state-of-the-art resides on attention-based models, with contextualized word representations being fine-tuned for temporal relation extraction. However, further experiments and advances in the research topic are required until real-time clinical domain applications are released. Furthermore, most of the publications mainly reside on the same dataset, hindering the need for new annotation projects that provide datasets for different medical specialties, clinical text types, and even languages.

Detecting formal thought disorder by deep contextualized word representations

Computational linguistics has enabled the introduction of objective tools that measure some of the symptoms of schizophrenia, including the coherence of speech associated with formal thought disorder (FTD). Our goal was to investigate whether neural network based utterance embeddings are more accurate in detecting FTD than models based on individual indicators. The present research used a comprehensive Embeddings from Language Models (ELMo) approach to represent interviews with patients suffering from schizophrenia (N=35) and with healthy people (N=35). We compared its results to the approach described by Bedi et al. (2015), referred to here as the coherence model. Evaluations were also performed by a clinician using the Scale for the Assessment of Thought, Language and Communication (TLC). Using all six TLC questions the ELMo obtained an accuracy of 80% in distinguishing patients from healthy people. Previously used coherence models were less accurate at 70%. The classifying clinician was accurate 74% of the time. Our analysis shows that both ELMo and TLC are sensitive to the symptoms of disorganization in patients. In this study methods using text representations from language models were more accurate than those based solely on the assessment of FTD, and can be used as measures of disordered language that complement human clinical ratings.

Relation Classification for Bleeding Events From Electronic Health Records Using Deep Learning Systems: An Empirical Study

BackgroundAccurate detection of bleeding events from electronic health records (EHRs) is crucial for identifying and characterizing different common and serious medical problems. To extract such information from EHRs, it is essential to identify the relations between bleeding events and related clinical entities (eg, bleeding anatomic sites and lab tests). With the advent of natural language processing (NLP) and deep learning (DL)-based techniques, many studies have focused on their applicability for various clinical applications. However, no prior work has utilized DL to extract relations between bleeding events and relevant entities.ObjectiveIn this study, we aimed to evaluate multiple DL systems on a novel EHR data set for bleeding event–related relation classification.MethodsWe first expert annotated a new data set of 1046 deidentified EHR notes for bleeding events and their attributes. On this data set, we evaluated three state-of-the-art DL architectures for the bleeding event relation classification task, namely, convolutional neural network (CNN), attention-guided graph convolutional network (AGGCN), and Bidirectional Encoder Representations from Transformers (BERT). We used three BERT-based models, namely, BERT pretrained on biomedical data (BioBERT), BioBERT pretrained on clinical text (Bio+Clinical BERT), and BioBERT pretrained on EHR notes (EhrBERT).ResultsOur experiments showed that the BERT-based models significantly outperformed the CNN and AGGCN models. Specifically, BioBERT achieved a macro F1 score of 0.842, outperforming both the AGGCN (macro F1 score, 0.828) and CNN models (macro F1 score, 0.763) by 1.4% (P<.001) and 7.9% (P<.001), respectively.ConclusionsIn this comprehensive study, we explored and compared different DL systems to classify relations between bleeding events and other medical concepts. On our corpus, BERT-based models outperformed other DL models for identifying the relations of bleeding-related entities. In addition to pretrained contextualized word representation, BERT-based models benefited from the use of target entity representation over traditional sequence representation

ActTRANS: Functional classification in active transport proteins based on transfer learning and contextual representations

MotivationPrimary and secondary active transport are two types of active transport that involve using energy to move the substances. Active transport mechanisms do use proteins to assist in transport and play essential roles to regulate the traffic of ions or small molecules across a cell membrane against the concentration gradient. In this study, the two main types of proteins involved in such transport are classified from transmembrane transport proteins. We propose a Support Vector Machine (SVM) with contextualized word embeddings from Bidirectional Encoder Representations from Transformers (BERT) to represent protein sequences. BERT is a powerful model in transfer learning, a deep learning language representation model developed by Google and one of the highest performing pre-trained model for Natural Language Processing (NLP) tasks. The idea of transfer learning with pre-trained model from BERT is applied to extract fixed feature vectors from the hidden layers and learn contextual relations between amino acids in the protein sequence. Therefore, the contextualized word representations of proteins are introduced to effectively model complex structures of amino acids in the sequence and the variations of these amino acids in the context. By generating context information, we capture multiple meanings for the same amino acid to reveal the importance of specific residues in the protein sequence. ResultsThe performance of the proposed method is evaluated using five-fold cross-validation and independent test. The proposed method achieves an accuracy of 85.44 %, 88.74 % and 92.84 % for Class-1, Class-2, and Class-3, respectively. Experimental results show that this approach can outperform from other feature extraction methods using context information, effectively classify two types of active transport and improve the overall performance.

Co-GAT: A Co-Interactive Graph Attention Network for Joint Dialog Act Recognition and Sentiment Classification

In a dialog system, dialog act recognition and sentiment classification are two correlative tasks to capture speakers’ intentions, where dialog act and sentiment can indicate the explicit and the implicit intentions separately. The dialog context information (contextual information) and the mutual interaction information are two key factors that contribute to the two related tasks. Unfortunately, none of the existing approaches consider the two important sources of information simultaneously. In this paper, we propose a Co-Interactive Graph Attention Network (Co-GAT) to jointly perform the two tasks. The core module is a proposed co-interactive graph interaction layer where a cross-utterances connection and a cross-tasks connection are constructed and iteratively updated with each other, achieving to consider the two types of information simultaneously. Experimental results on two public datasets show that our model successfully captures the two sources of information and achieve the state-of-the-art performance. In addition, we find that the contributions from the contextual and mutual interaction information do not fully overlap with contextualized word representations (BERT, Roberta, XLNet).

Filling the Gap of Utterance-aware and Speaker-aware Representation for Multi-turn Dialogue

A multi-turn dialogue is composed of multiple utterances from two or more different speaker roles. Thus utterance- and speaker-aware clues are supposed to be well captured in models. However, in the existing retrieval-based multi-turn dialogue modeling, the pre-trained language models (PrLMs) as encoder represent the dialogues coarsely by taking the pairwise dialogue history and candidate response as a whole, the hierarchical information on either utterance interrelation or speaker roles coupled in such representations is not well addressed. In this work, we propose a novel model to fill such a gap by modeling the effective utterance-aware and speaker-aware representations entailed in a dialogue history. In detail, we decouple the contextualized word representations by masking mechanisms in Transformer-based PrLM, making each word only focus on the words in current utterance, other utterances, two speaker roles (i.e., utterances of sender and utterances of receiver), respectively. Experimental results show that our method boosts the strong ELECTRA baseline substantially in four public benchmark datasets, and achieves various new state-of-the-art performance over previous methods. A series of ablation studies are conducted to demonstrate the effectiveness of our method.

Biomedical event trigger extraction based on multi-layer residual BiLSTM and contextualized word representations

Biomedical event extraction is an important branch of biomedical information extraction. Trigger extraction is the most essential sub-task in event extraction, which has been widely concerned. Existing trigger extraction studies are mostly based on conventional machine learning or neural networks. But they neglect the ambiguity of word representations and the insufficient feature extraction by shallow hidden layers. In this paper, trigger extraction is treated as a sequence labeling problem. We introduce the language model to dynamically compute contextualized word representations and propose a multi-layer residual bidirectional long short-term memory (BiLSTM) architecture. First, we concatenate contextualized word embedding, pretrained word embedding and character-level embedding as the feature representations, which effectively solves the tokens’ ambiguity in biomedical corpora. Then, the designed BiLSTM block with residual connection and gated multi-layer perceptron is adopted to extract features iteratively. This architecture improves the ability of our model to capture information and avoids gradient exploding or vanishing. Finally, we combine the multi-layer residual BiLSTM with CRF layer to obtain more reasonable label sequences. Comparing with other state-of-the-art methods, the proposed model achieves the competitive performance (F1-score: 80.74%) on the biomedical multi-level event extraction (MLEE) corpus without any manual participation and feature engineering.