• The Journal of Bigdata
• /
• v.7 no.2
• /
• pp.11-29
• /
• 2022

Pre-trained language models are the most important and widely used tools in natural language processing tasks. Since those have been pre-trained for a large amount of corpus, high performance can be expected even with fine-tuning learning using a small number of data. Since the elements necessary for implementation, such as a pre-trained tokenizer and a deep learning model including pre-trained weights, are distributed together, the cost and period of natural language processing has been greatly reduced. Transformer variants are the most representative pre-trained language models that provide these advantages. Those are being actively used in other fields such as computer vision and audio applications. In order to make it easier for researchers to understand the pre-trained language model and apply it to natural language processing tasks, this paper describes the definition of the language model and the pre-learning language model, and discusses the development process of the pre-trained language model and especially representative Transformer variants.

• Journal of Information Technology Services
• /
• v.19 no.5
• /
• pp.83-91
• /
• 2020

Language Models such as BERT has been an important factor of deep learning-based natural language processing. Pre-training the transformer-based language models would be computationally expensive since they are consist of deep and broad architecture and layers using an attention mechanism and also require huge amount of data to train. Hence, it became mandatory to do fine-tuning large pre-trained language models which are trained by Google or some companies can afford the resources and cost. There are various techniques for fine tuning the language models and this paper examines three techniques, which are data augmentation, tuning the hyper paramters and partly re-constructing the neural networks. For data augmentation, we use no-answer augmentation and back-translation method. Also, some useful combinations of hyper parameters are observed by conducting a number of experiments. Finally, we have GRU, LSTM networks to boost our model performance with adding those networks to BERT pre-trained model. We do fine-tuning the pre-trained korean-based language model through the methods mentioned above and push the F1 score from baseline up to 89.66. Moreover, some failure attempts give us important lessons and tell us the further direction in a good way.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.2
• /
• pp.15-23
• /
• 2022

In the field of patents, as NLP(Natural Language Processing) is a challenging task due to the linguistic specificity of patent literature, there is an urgent need to research a language model optimized for Korean patent literature. Recently, in the field of NLP, there have been continuous attempts to establish a pre-trained language model for specific domains to improve performance in various tasks of related fields. Among them, ELECTRA is a pre-trained language model by Google using a new method called RTD(Replaced Token Detection), after BERT, for increasing training efficiency. The purpose of this paper is to propose KorPatELECTRA pre-trained on a large amount of Korean patent literature data. In addition, optimal pre-training was conducted by preprocessing the training corpus according to the characteristics of the patent literature and applying patent vocabulary and tokenizer. In order to confirm the performance, KorPatELECTRA was tested for NER(Named Entity Recognition), MRC(Machine Reading Comprehension), and patent classification tasks using actual patent data, and the most excellent performance was verified in all the three tasks compared to comparative general-purpose language models.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.9
• /
• pp.2334-2347
• /
• 2023

Metaverse services generate text data, data of ubiquitous computing, in real-time to analyze user emotions. Analysis of user emotions is an important task in metaverse services. This study aims to classify user sentiments using deep learning and pre-trained language models based on the transformer structure. Previous studies collected data from a single platform, whereas the current study incorporated the review data as "Metaverse" keyword from the YouTube and Google Play Store platforms for general utilization. As a result, the Bidirectional Encoder Representations from Transformers (BERT) and Robustly optimized BERT approach (RoBERTa) models using the soft voting mechanism achieved a highest accuracy of 88.57%. In addition, the area under the curve (AUC) score of the ensemble model comprising RoBERTa, BERT, and A Lite BERT (ALBERT) was 0.9458. The results demonstrate that the ensemble combined with the RoBERTa model exhibits good performance. Therefore, the RoBERTa model can be applied on platforms that provide metaverse services. The findings contribute to the advancement of natural language processing techniques in metaverse services, which are increasingly important in digital platforms and virtual environments. Overall, this study provides empirical evidence that sentiment analysis using deep learning and pre-trained language models is a promising approach to improving user experiences in metaverse services.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.3
• /
• pp.771-791
• /
• 2022

Protein-protein interaction (PPI) extraction from original text is important for revealing the molecular mechanism of biological processes. With the rapid growth of biomedical literature, manually extracting PPI has become more time-consuming and laborious. Therefore, the automatic PPI extraction from the raw literature through natural language processing technology has attracted the attention of the majority of researchers. We propose a PPI extraction model based on the large pre-trained language model and adversarial training. It enhances the learning of semantic and syntactic features using BioBERT pre-trained weights, which are built on large-scale domain corpora, and adversarial perturbations are applied to the embedding layer to improve the robustness of the model. Experimental results showed that the proposed model achieved the highest F1 scores (83.93% and 90.31%) on two corpora with large sample sizes, namely, AIMed and BioInfer, respectively, compared with the previous method. It also achieved comparable performance on three corpora with small sample sizes, namely, HPRD50, IEPA, and LLL.

• Journal of Intelligence and Information Systems
• /
• v.28 no.4
• /
• pp.251-273
• /
• 2022

Recently, research on applying text analysis to deep learning has steadily continued. In particular, researches have been actively conducted to understand the meaning of words and perform tasks such as summarization and sentiment classification through a pre-trained language model that learns large datasets. However, existing pre-trained language models show limitations in that they do not understand specific domains well. Therefore, in recent years, the flow of research has shifted toward creating a language model specialized for a particular domain. Domain-specific pre-trained language models allow the model to understand the knowledge of a particular domain better and reveal performance improvements on various tasks in the field. However, domain-specific further pre-training is expensive to acquire corpus data of the target domain. Furthermore, many cases have reported that performance improvement after further pre-training is insignificant in some domains. As such, it is difficult to decide to develop a domain-specific pre-trained language model, while it is not clear whether the performance will be improved dramatically. In this paper, we present a way to proactively check the expected performance improvement by further pre-training in a domain before actually performing further pre-training. Specifically, after selecting three domains, we measured the increase in classification accuracy through further pre-training in each domain. We also developed and presented new indicators to estimate the specificity of the domain based on the normalized frequency of the keywords used in each domain. Finally, we conducted classification using a pre-trained language model and a domain-specific pre-trained language model of three domains. As a result, we confirmed that the higher the domain specificity index, the higher the performance improvement through further pre-training.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.6
• /
• pp.1692-1705
• /
• 2024

In this paper, we present a method that integrates a Grammar Transducer as an external language model to enhance the accuracy of the pre-trained Korean End-to-end (E2E) Automatic Speech Recognition (ASR) model. The E2E ASR model utilizes the Connectionist Temporal Classification (CTC) loss function to derive hypothesis sentences from input audio. However, this method reveals a limitation inherent in the CTC approach, as it fails to capture language information from transcript data directly. To overcome this limitation, we propose a fusion approach that combines a clause-level n-gram language model, transformed into a Weighted Finite-State Transducer (WFST), with the E2E ASR model. This approach enhances the model's accuracy and allows for domain adaptation using just additional text data, avoiding the need for further intensive training of the extensive pre-trained ASR model. This is particularly advantageous for Korean, characterized as a low-resource language, which confronts a significant challenge due to limited resources of speech data and available ASR models. Initially, we validate the efficacy of training the n-gram model at the clause-level by contrasting its inference accuracy with that of the E2E ASR model when merged with language models trained on smaller lexical units. We then demonstrate that our approach achieves enhanced domain adaptation accuracy compared to Shallow Fusion, a previously devised method for merging an external language model with an E2E ASR model without necessitating additional training.

• Journal of Korea Society of Industrial Information Systems
• /
• v.29 no.3
• /
• pp.79-87
• /
• 2024

Recently, the importance of pre-trained language models (PLM) has been emphasized for natural language processing (NLP) such as text classification, sentiment analysis, and question answering. Korean PLM shows high performance in NLP in general-purpose domains, but is weak in domains such as finance, medicine, and law. The main goal of this study is to propose a language model learning process and method to build a financial-specific language model that shows good performance not only in the financial domain but also in general-purpose domains. The five steps of the financial-specific language model are (1) financial data collection and preprocessing, (2) selection of model architecture such as PLM or foundation model, (3) domain data learning and instruction tuning, (4) model verification and evaluation, and (5) model deployment and utilization. Through this, a method for constructing pre-learning data that takes advantage of the characteristics of the financial domain and an efficient LLM training method, adaptive learning and instruction tuning techniques, were presented.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.10
• /
• pp.157-165
• /
• 2021

Recently, studies using deep learning to analyze a large amount of text are being actively conducted. In particular, a pre-trained language model that applies the learning results of a large amount of text to the analysis of a specific domain text is attracting attention. Among various pre-trained language models, BERT(Bidirectional Encoder Representations from Transformers)-based model is the most widely used. Recently, research to improve the performance of analysis is being conducted through further pre-training using BERT's MLM(Masked Language Model). However, the traditional MLM has difficulties in clearly understands the meaning of sentences containing new words such as newly coined words. Therefore, in this study, we newly propose NTM(Newly coined words Target Masking), which performs masking only on new words. As a result of analyzing about 700,000 movie reviews of portal 'N' by applying the proposed methodology, it was confirmed that the proposed NTM showed superior performance in terms of accuracy of sensitivity analysis compared to the existing random masking.

• Journal of the Korea Society of Computer and Information
• /
• v.25 no.5
• /
• pp.187-197
• /
• 2020

Recently, various methods of text embedding using deep learning algorithms have been proposed. Especially, the way of using pre-trained language model which uses tremendous amount of text data in training is mainly applied for embedding new text data. However, traditional pre-trained language model has some limitations that it is hard to understand unique context of new text data when the text has too many tokens. In this paper, we propose self-supervised learning-based fine tuning method for pre-trained language model to infer vectors of long-text. Also, we applied our method to news articles and classified them into categories and compared classification accuracy with traditional models. As a result, it was confirmed that the vector generated by the proposed model more accurately expresses the inherent characteristics of the document than the vectors generated by the traditional models.