• Journal of the Korea Society of Computer and Information
• /
• v.27 no.11
• /
• pp.39-46
• /
• 2022

In this paper, we propose a multi-task model that can simultaneously predict general-purpose tasks such as part-of-speech tagging, lemmatization, and dependency parsing using the UD Korean Kaist v2.3 corpus. The proposed model thus applies the self-attention technique of the BERT model and the graph-based Biaffine attention technique by fine-tuning the multilingual BERT and the two Korean-specific BERTs such as KR-BERT and KoBERT. The performances of the proposed model are compared and analyzed using the multilingual version of BERT and the two Korean-specific BERT language models.

• Knowledge Management Research
• /
• v.22 no.3
• /
• pp.91-106
• /
• 2021

With the recent rapid development of deep learning technology, the demand for analyzing huge text documents in the national R&D field from various perspectives is rapidly increasing. In particular, interest in the application of a BERT(Bidirectional Encoder Representations from Transformers) language model that has pre-trained a large corpus is growing. However, the terminology used frequently in highly specialized fields such as national R&D are often not sufficiently learned in basic BERT. This is pointed out as a limitation of understanding documents in specialized fields through BERT. Therefore, this study proposes a method to build an R&D KoBERT language model that transfers national R&D field knowledge to basic BERT using further pre-training. In addition, in order to evaluate the performance of the proposed model, we performed classification analysis on about 116,000 R&D reports in the health care and information and communication fields. Experimental results showed that our proposed model showed higher performance in terms of accuracy compared to the pure KoBERT model.

• Journal of Advanced Information Technology and Convergence
• /
• v.10 no.1
• /
• pp.45-56
• /
• 2020

Despite extensive research, performance enhancement of keyphrase (KP) extraction remains a challenging problem in modern informatics. Recently, deep learning-based supervised approaches have exhibited state-of-the-art accuracies with respect to this problem, and several of the previously proposed methods utilize Bidirectional Encoder Representations from Transformers (BERT)-based language models. However, few studies have investigated the effective application of BERT-based fine-tuning techniques to the problem of KP extraction. In this paper, we consider the aforementioned problem in the context of scientific articles by investigating the fine-tuning characteristics of two distinct BERT models - BERT (i.e., base BERT model by Google) and SciBERT (i.e., a BERT model trained on scientific text). Three different datasets (WWW, KDD, and Inspec) comprising data obtained from the computer science domain are used to compare the results obtained by fine-tuning BERT and SciBERT in terms of KP extraction.

• Journal of Internet Computing and Services
• /
• v.24 no.4
• /
• pp.127-135
• /
• 2023

This research paper explores the application of FinBERT, a variational BERT-based model pre-trained on financial domain, for sentiment analysis in the financial domain while focusing on the process of identifying suitable training data and hyperparameters. Our goal is to offer a comprehensive guide on effectively utilizing the FinBERT model for accurate sentiment analysis by employing various datasets and fine-tuning hyperparameters. We outline the architecture and workflow of the proposed approach for fine-tuning the FinBERT model in this study, emphasizing the performance of various datasets and hyperparameters for sentiment analysis tasks. Additionally, we verify the reliability of GPT-3 as a suitable annotator by using it for sentiment labeling tasks. Our results show that the fine-tuned FinBERT model excels across a range of datasets and that the optimal combination is a learning rate of 5e-5 and a batch size of 64, which perform consistently well across all datasets. Furthermore, based on the significant performance improvement of the FinBERT model with our Twitter data in general domain compared to our news data in general domain, we also express uncertainty about the model being further pre-trained only on financial news data. We simplify the complex process of determining the optimal approach to the FinBERT model and provide guidelines for selecting additional training datasets and hyperparameters within the fine-tuning process of financial sentiment analysis models.

• Electronics and Telecommunications Trends
• /
• v.35 no.3
• /
• pp.9-19
• /
• 2020

Recently, a technique for applying a deep learning language model pretrained from a large corpus to fine-tuning for each application task has been widely used as a language processing technology. The pretrained language model shows higher performance and satisfactory generalization performance than existing methods. This paper introduces the major research trends related to deep learning pretrained language models in the field of language processing. We describe in detail the motivations, models, learning methods, and results of the BERT language model that had significant influence on subsequent studies. Subsequently, we introduce the results of language model studies after BERT, focusing on SpanBERT, RoBERTa, ALBERT, BART, and ELECTRA. Finally, we introduce the KorBERT pretrained language model, which shows satisfactory performance in Korean language. In addition, we introduce techniques on how to apply the pretrained language model to Korean (agglutinative) language, which consists of a combination of content and functional morphemes, unlike English (refractive) language whose endings change depending on the application.

• ETRI Journal
• /
• v.46 no.1
• /
• pp.35-47
• /
• 2024

This study introduces CR-M-SpanBERT, a coreference resolution (CR) model that utilizes multiple embedding-based span bidirectional encoder representations from transformers, for antecedent recognition in natural language (NL) text. Information extraction studies aimed to extract knowledge from NL text autonomously and cost-effectively. However, the extracted information may not represent knowledge accurately owing to the presence of ambiguous entities. Therefore, we propose a CR model that identifies mentions referring to the same entity in NL text. In the case of CR, it is necessary to understand both the syntax and semantics of the NL text simultaneously. Therefore, multiple embeddings are generated for CR, which can include syntactic and semantic information for each word. We evaluate the effectiveness of CR-M-SpanBERT by comparing it to a model that uses SpanBERT as the language model in CR studies. The results demonstrate that our proposed deep neural network model achieves high-recognition accuracy for extracting antecedents from NL text. Additionally, it requires fewer epochs to achieve an average F1 accuracy greater than 75% compared with the conventional SpanBERT approach.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.437-440
• /
• 2020

최근 자연어 처리에서는 사전 학습과 전이 학습을 통하여 다양한 과제에 높은 성능 향상을 성취하고 있다. 사전 학습의 대표적 모델로 구글의 BERT가 있으며, 구글에서 제공한 다국어 모델을 포함해 한국의 여러 연구기관과 기업에서 한국어 데이터셋으로 학습한 BERT 모델을 제공하고 있다. 하지만 이런 BERT 모델들은 사전 학습에 사용한 말뭉치의 특성에 따라 이후 전이 학습에서의 성능 차이가 발생한다. 본 연구에서는 소셜미디어에서 나타나는 구어체와 신조어, 특수문자, 이모지 등 일반 사용자들의 문장에 보다 유연하게 대응할 수 있는 한국어 뉴스 댓글 데이터를 통해 학습한 KcBERT를 소개한다. 본 모델은 최소한의 데이터 정제 이후 BERT WordPiece 토크나이저를 학습하고, BERT Base 모델과 BERT Large 모델을 모두 학습하였다. 또한, 학습된 모델을 HuggingFace Model Hub에 공개하였다. KcBERT를 기반으로 전이 학습을 통해 한국어 데이터셋에 적용한 성능을 비교한 결과, 한국어 영화 리뷰 코퍼스(NSMC)에서 최고 성능의 스코어를 얻을 수 있었으며, 여타 데이터셋에서는 기존 한국어 BERT 모델과 비슷한 수준의 성능을 보였다.

• Phonetics and Speech Sciences
• /
• v.15 no.3
• /
• pp.69-74
• /
• 2023

This study introduces an innovative model for zero-shot voice conversion that utilizes the capabilities of HuBERT. Zero-shot voice conversion models can transform the speech of one speaker to mimic that of another, even when the model has not been exposed to the target speaker's voice during the training phase. Comprising five main components (HuBERT, feature encoder, flow, speaker encoder, and vocoder), the model offers remarkable performance across a range of scenarios. Notably, it excels in the challenging unseen-to-unseen voice-conversion tasks. The effectiveness of the model was assessed based on the mean opinion scores and similarity scores, reflecting high voice quality and similarity to the target speakers. This model demonstrates considerable promise for a range of real-world applications demanding high-quality voice conversion. This study sets a precedent in the exploration of HuBERT-based models for voice conversion, and presents new directions for future research in this domain. Despite its complexities, the robust performance of this model underscores the viability of HuBERT in advancing voice conversion technology, making it a significant contributor to the field.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.3-7
• /
• 2019

자연어 표상은 자연어가 가진 정보를 컴퓨터에게 전달하기 위해 표현하는 방법이다. 현재 자연어 표상은 학습을 통해 고정된 벡터로 표현하는 것이 아닌 문맥적 정보에 의해 벡터가 변화한다. 그 중 BERT의 경우 Transformer 모델의 encoder를 사용하여 자연어를 표상하는 기술이다. 하지만 BERT의 경우 학습시간이 많이 걸리며, 대용량의 데이터를 필요로 한다. 본 논문에서는 빠른 자연어 표상 학습을 위해 의미 정보와 BERT를 결합한 개념 언어 모델을 제안한다. 의미 정보로 단어의 품사 정보와, 명사의 의미 계층 정보를 추상적으로 표현했다. 실험을 위해 ETRI에서 공개한 한국어 BERT 모델을 비교 대상으로 하며, 개체명 인식을 학습하여 비교했다. 두 모델의 개체명 인식 결과가 비슷하게 나타났다. 의미 정보가 자연어 표상을 하는데 중요한 정보가 될 수 있음을 확인했다.

• KIPS Transactions on Software and Data Engineering
• /
• v.11 no.4
• /
• pp.169-178
• /
• 2022

Morphemes are most primitive units in a language that lose their original meaning when segmented into smaller parts. In Korean, a sentence is a sequence of eojeols (words) separated by spaces. Each eojeol comprises one or more morphemes. Korean morphological analysis (KMA) is to divide eojeols in a given Korean sentence into morpheme units. It also includes assigning appropriate part-of-speech(POS) tags to the resulting morphemes. KMA is one of the most important tasks in Korean natural language processing (NLP). Improving the performance of KMA is closely related to increasing performance of Korean NLP tasks. Recent research on KMA has begun to adopt the approach of machine translation (MT) models. MT is to convert a sequence (sentence) of units of one domain into a sequence (sentence) of units of another domain. Neural machine translation (NMT) stands for the approaches of MT that exploit neural network models. From a perspective of MT, KMA is to transform an input sequence of units belonging to the eojeol domain into a sequence of units in the morpheme domain. In this paper, we propose a deep learning model for KMA. The backbone of our model is based on the BERT-fused model which was shown to achieve high performance on NMT. The BERT-fused model utilizes Transformer, a representative model employed by NMT, and BERT which is a language representation model that has enabled a significant advance in NLP. The experimental results show that our model achieves 98.24 F1-Score.