• Journal of Intelligence and Information Systems
• /
• v.28 no.4
• /
• pp.309-328
• /
• 2022

This study empirically analyzed a Korean pre-trained language models (PLMs) designed for natural language generation. The performance of two PLMs - BART and GPT - at the task of abstractive text summarization was compared. To investigate how performance depends on the characteristics of the inference data, ten different document types, containing six types of informational content and creation content, were considered. It was found that BART (which can both generate and understand natural language) performed better than GPT (which can only generate). Upon more detailed examination of the effect of inference data characteristics, the performance of GPT was found to be proportional to the length of the input text. However, even for the longest documents (with optimal GPT performance), BART still out-performed GPT, suggesting that the greatest influence on downstream performance is not the size of the training data or PLMs parameters but the structural suitability of the PLMs for the applied downstream task. The performance of different PLMs was also compared through analyzing parts of speech (POS) shares. BART's performance was inversely related to the proportion of prefixes, adjectives, adverbs and verbs but positively related to that of nouns. This result emphasizes the importance of taking the inference data's characteristics into account when fine-tuning a PLMs for its intended downstream task.

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.4
• /
• pp.159-172
• /
• 2023

In general, social problem-solving research aims to create important social value by offering meaningful answers to various social pending issues using scientific technologies. Not surprisingly, however, although numerous and extensive research attempts have been made to alleviate the social problems and issues in nation-wide, we still have many important social challenges and works to be done. In order to facilitate the entire process of the social problem-solving research and maximize its efficacy, it is vital to clearly identify and grasp the important and pressing problems to be focused upon. It is understandable for the problem discovery step to be drastically improved if current social issues can be automatically identified from existing R&D resources such as technical reports and articles. This paper introduces a comprehensive dataset which is essential to build a machine learning model for automatically detecting the social problems and solutions in various national research reports. Initially, we collected a total of 700 research reports regarding social problems and issues. Through intensive annotation process, we built totally 24,022 sentences each of which possesses its own category or label closely related to social problem-solving such as problems, purposes, solutions, effects and so on. Furthermore, we implemented four sentence classification models based on various neural language models and conducted a series of performance experiments using our dataset. As a result of the experiment, the model fine-tuned to the KLUE-BERT pre-trained language model showed the best performance with an accuracy of 75.853% and an F1 score of 63.503%.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.507-508
• /
• 2021

최근 한국어 자연어처리 과제에서 대형 언어 모델을 통해 다양한 언어처리 작업에 대한 연구가 활발히 이루어지고 있다. 특히 동형이의어를 구분하는 작업은 문장의 문법성을 정확히 판단하고 비교해야 되기 때문에 어려운 작업이다. KE-T5는 큰 규모의 한국어를 통해 학습된 한국어 대형 언어 모델로 대부분의 자연어처리 과제에서 활용할 수 있으며 복잡한 언어처리 작업에서 높은 성능을 기대할 수 있다. 본 논문에서는 큰 규모의 한국어를 통해 학습된 KE-T5를 활용하여 동형이의어 구별 문제를 수행하고 평가한다.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.505-506
• /
• 2021

최근 인터넷에서 기하급수적으로 증가하는 방대한 양의 텍스트를 자동으로 요약하려는 연구가 활발하게 이루어지고 있다. 자동 텍스트 요약 작업은 다양한 사전학습 모델의 등장으로 인해 많은 발전을 이루었다. 특히 T5(Text-to-Text Transfer Transformer) 기반의 모델은 자동 텍스트 요약 작업에서 매우 우수한 성능을 보이며, 해당 분야의 SOTA(State of the Art)를 달성하고 있다. 본 논문에서는 방대한 양의 한국어를 학습시킨 사전학습 모델 KE-T5를 활용하여 자동 논문 요약을 수행하고 평가한다.

• International Journal of Computer Science & Network Security
• /
• v.23 no.4
• /
• pp.15-24
• /
• 2023

One of the objectives and aspirations of scientists and engineers ever since the development of computers has been to interact naturally with machines. Hence features of artificial intelligence (AI) like natural language processing and natural language generation were developed. The field of AI that is thought to be expanding the fastest is interactive conversational systems. Numerous businesses have created various Virtual Personal Assistants (VPAs) using these technologies, including Apple's Siri, Amazon's Alexa, and Google Assistant, among others. Even though many chatbots have been introduced through the years to diagnose or treat psychological disorders, we are yet to have a user-friendly chatbot available. A smart generative cognitive behavioral therapy with spoken dialogue systems support was then developed using a model Persona Perception (P2) bot with Generative Pre-trained Transformer-2 (GPT-2). The model was then implemented using modern technologies in VPAs like voice recognition, Natural Language Understanding (NLU), and text-to-speech. This system is a magnificent device to help with voice-based systems because it can have therapeutic discussions with the users utilizing text and vocal interactive user experience.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2020.11a
• /
• pp.1014-1017
• /
• 2020

Many natural language processing (NLP) models utilize pre-trained word embeddings to leverage latent information. One of the most successful word embedding model is the Skip-gram (SG). In this paper, we propose a Skipgram drop (SG-Drop) model, which is a variation of the SG model. The SG-Drop model is designed to reduce training time efficiently. Furthermore, the SG-Drop allows controlling training time with its hyperparameter. It could train word embedding faster than reducing training epochs while better preserving the quality.

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.8
• /
• pp.371-380
• /
• 2023

Relation extraction is to extract relationships between named entities from text. Traditionally, relation extraction methods only extract relations between predetermined subject and object entities. However, in end-to-end relation extraction, all possible relations must be extracted by considering the positions of the subject and object for each pair of entities, and so this method uses time and resources inefficiently. To alleviate this problem, this paper proposes a method that sets directions based on the positions of the subject and object, and extracts relations according to the directions. The proposed method utilizes existing relation extraction data to generate direction labels indicating the direction in which the subject points to the object in the sentence, adds entity position tokens and entity type to sentences to predict the directions using a pre-trained language model (KLUE-RoBERTa-base, RoBERTa-base), and generates representations of subject and object entities through probabilistic crossover operation. Then, we make use of these representations to extract relations. Experimental results show that the proposed model performs about 3 ~ 4%p better than a method for predicting integrated labels. In addition, when learning Korean and English data using the proposed model, the performance was 1.7%p higher in English than in Korean due to the number of data and language disorder and the values of the parameters that produce the best performance were different. By excluding the number of directional cases, the proposed model can reduce the waste of resources in end-to-end relation extraction.

• Journal of the Korea Convergence Society
• /
• v.13 no.2
• /
• pp.13-20
• /
• 2022

This study is to develop a named entity recognition model specialized in criminal investigation domains using deep learning techniques. Through this study, we propose a system that can contribute to analysis of crime for prevention and investigation using data analysis techniques in the future by automatically extracting and categorizing crime-related information from text-based data such as criminal judgments and investigation documents. For this study, the criminal investigation domain text was collected and the required entity name was newly defined from the perspective of criminal analysis. In addition, the proposed model applying KoELECTRA, a pre-trained language model that has recently shown high performance in natural language processing, shows performance of micro average(referred to as micro avg) F1-score 98% and macro average(referred to as macro avg) F1-score 95% in 9 main categories of crime domain NER experiment data, and micro avg F1-score 98% and macro avg F1-score 62% in 56 sub categories. The proposed model is analyzed from the perspective of future improvement and utilization.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.18 no.4
• /
• pp.693-700
• /
• 2023

Sentiment analysis, a branch of natural language processing that classifies and identifies subjective opinions and emotions in text documents as positive or negative, can be used for various promotions and services through customer preference analysis. To this end, recent research has been conducted utilizing various techniques in machine learning and deep learning. In this study, we propose an optimal language model by comparing the accuracy of sentiment analysis for movie, product, and game reviews using existing RNN-based models and recent Transformer-based language models. In our experiments, LMKorBERT and GPT3 showed relatively good accuracy among the models pre-trained on the Korean corpus.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.10a
• /
• pp.164-167
• /
• 2022

In table Machine comprehension, the knowledge required for language models or the structural form of tables changes depending on the domain, showing a greater performance degradation compared to text data. In this paper, we propose a pre-learning data construction method and an adversarial learning method through meaningful tabular data selection for constructing a pre-learning table language model robust to these domain changes in table machine reading. In order to detect tabular data sed for decoration of web documents without structural information from the extracted table data, a rule through heuristic was defined to identify head data and select table data was applied. An adversarial learning method between tabular data and infobax data with knowledge information about entities was applied. When the data was refined compared to when it was trained with the existing unrefined data, F1 3.45 and EM 4.14 increased in the KorQuAD table data, and F1 19.38, EM 4.22 compared to when the data was not refined in the Spec table QA data showed increased performance.