• Smart Media Journal
• /
• v.12 no.1
• /
• pp.32-46
• /
• 2023

Natural Language Processing (NLP) has grown tremendously in recent years. Typically, bilingual, and multilingual translation models have been deployed widely in machine translation and gained vast attention from the research community. On the contrary, few studies have focused on translating between spoken and sign languages, especially non-English languages. Prior works on Sign Language Translation (SLT) have shown that a mid-level sign gloss representation enhances translation performance. Therefore, this study presents a new large-scale Korean sign language dataset, the Museum-Commentary Korean Sign Gloss (MCKSG) dataset, including 3828 pairs of Korean sentences and their corresponding sign glosses used in Museum-Commentary contexts. In addition, we propose a translation framework based on self-supervised learning, where the pretext task is a text-to-text from a Korean sentence to its back-translation versions, then the pre-trained network will be fine-tuned on the MCKSG dataset. Using self-supervised learning help to overcome the drawback of a shortage of sign language data. Through experimental results, our proposed model outperforms a baseline BERT model by 6.22%.

• The Journal of Information Systems
• /
• v.32 no.4
• /
• pp.119-136
• /
• 2023

Purpose We introduce a groundbreaking surveillance system explicitly designed to overcome the limitations typically associated with conventional surveillance systems, which often focus primarily on object-centric behavior analysis. Design/methodology/approach The study introduces an innovative approach to risk assessment in surveillance, employing image captioning to generate descriptive captions that effectively encapsulate the interactions among objects, actions, and spatial elements within observed scenes. To support our methodology, we developed a distinctive dataset comprising pairs of [image-caption-danger score] for training purposes. We fine-tuned the BLIP-2 model using this dataset and utilized BERT to decipher the semantic content of the generated captions for assessing risk levels. Findings In a series of experiments conducted with our self-constructed datasets, we illustrate that these datasets offer a wealth of information for risk assessment and display outstanding performance in this area. In comparison to models pre-trained on established datasets, our generated captions thoroughly encompass the necessary object attributes, behaviors, and spatial context crucial for the surveillance system. Additionally, they showcase adaptability to novel sentence structures, ensuring their versatility across a range of contexts.

• Journal of the Korean Society for information Management
• /
• v.39 no.3
• /
• pp.99-132
• /
• 2022

The purpose of this study is to propose AET (Augmented and Extended Topics), a novel method of synthesizing both LDA and BERTopic results, and to analyze the recently published LIS articles as an experimental approach. To achieve the purpose of this study, 55,442 abstracts from 85 LIS journals within the WoS database, which spans from January 2001 to October 2021, were analyzed. AET first constructs a WORD2VEC-based cosine similarity matrix between LDA and BERTopic results, extracts AT (Augmented Topics) by repeating the matrix reordering and segmentation procedures as long as their semantic relations are still valid, and finally determines ET (Extended Topics) by removing any LDA related residual subtopics from the matrix and ordering the rest of them by F₁ (BERTopic topic size rank, Inverse cosine similarity rank). AET, by comparing with the baseline LDA result, shows that AT has effectively concretized the original LDA topic model and ET has discovered new meaningful topics that LDA didn't. When it comes to the qualitative performance evaluation, AT performs better than LDA while ET shows similar performances except in a few cases.

• Smart Media Journal
• /
• v.11 no.10
• /
• pp.46-53
• /
• 2022

Today, due to the 4th industrial revolution and extensive R&D funding, domestic companies have begun to possess world-class industrial technologies and have grown into important assets. The national government has designated it as a "national core technology" in order to protect companies' critical industrial technologies. Particularly, technology leaks in the shipbuilding, display, and semiconductor industries can result in a significant loss of competitiveness not only at the company level but also at the national level. Every year, there are more insider leaks, ransomware attacks, and attempts to steal industrial technology through industrial spy. The stolen industrial technology is then traded covertly on the dark web. In this paper, we propose a system for detecting industrial technology leaks in the dark web environment. The proposed model first builds a database through dark web crawling using information collected from the OSINT environment. Afterwards, keywords for industrial technology leakage are extracted using the KeyBERT model, and signs of industrial technology leakage in the dark web environment are proposed as quantitative figures. Finally, based on the identified industrial technology leakage sites in the dark web environment, the possibility of secondary leakage is detected through the PageRank algorithm. The proposed method accepted for the collection of 27,317 unique dark web domains and the extraction of 15,028 nuclear energy-related keywords from 100 nuclear power patents. 12 dark web sites identified as a result of detecting secondary leaks based on the highest nuclear leak dark web sites.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.43 no.1
• /
• pp.109-117
• /
• 2023

Schedule delays present a major risk factor, as they can adversely affect construction projects, such as through increasing construction costs, claims from a client, and/or a decrease in construction quality due to trims to stages to catch up on lost time. Risk management has been conducted according to the importance and priority of schedule delay risk, but quantification of risk on the depth of schedule delay tends to be inadequate due to limitations in data collection. Therefore, this research used the BERT (Bidirectional Encoder Representations from Transformers) language model to convert the contents of aconstruction log, which comprised unstructured data, into WBS (Work Breakdown Structure)-based structured data, and to form a model of classification and quantification of risk. A process was applied to eight highway construction sites, and 75 cases of rain schedule delay risk were obtained from 8 out of 39 detailed work kinds. Through a K-S test, a significant probability distribution was derived for fourkinds of work, and the risk impact was compared. The process presented in this study can be used to derive various schedule delay risks in construction projects and to quantify their depth.

• 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.21 no.2
• /
• pp.109-126
• /
• 2022

The number and variety of products and services offered by companies have increased dramatically, providing customers with more choices to meet their needs. As a solution to this information overload problem, the provision of tailored services to individuals has become increasingly important, and the personalized recommender systems have been widely studied and used in both academia and industry. Existing recommender systems face important problems in practical applications. The most important problem is that it cannot clearly explain why it recommends these products. In recent years, some researchers have found that the explanation of recommender systems may be very useful. As a result, users are generally increasing conversion rates, satisfaction, and trust in the recommender system if it is explained why those particular items are recommended. Therefore, this study presents a methodology of providing an explanatory function of a recommender system using a review text left by a user. The basic idea is not to use all of the user's reviews, but to provide them in a summarized form using only reviews left by similar users or neighbors involved in recommending the item as an explanation when providing the recommended item to the user. To achieve this research goal, this study aims to provide a product recommendation list using user-based collaborative filtering techniques, combine reviews left by neighboring users with each product to build a model that combines text summary methods among deep learning-based natural language processing methods. Using the IMDb movie database, text reviews of all target user neighbors' movies are collected and summarized to present descriptions of recommended movies. There are several text summary methods, but this study aims to evaluate whether the review summary is well performed by training the Sequence-to-sequence+attention model, which is a representative generation summary method, and the BertSum model, which is an extraction summary model.

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.11
• /
• pp.471-480
• /
• 2023

The Korean construction order volume in South Korea grew significantly from 91.3 trillion won in public orders in 2013 to a total of 212 trillion won in 2021, particularly in the private sector. As the size of the domestic and overseas markets grew, the scale and complexity of EPC (Engineering, Procurement, Construction) projects increased, and risk management of project management and ITB (Invitation to Bid) documents became a critical issue. The time granted to actual construction companies in the bidding process following the EPC project award is not only limited, but also extremely challenging to review all the risk terms in the ITB document due to manpower and cost issues. Previous research attempted to categorize the risk terms in EPC contract documents and detect them based on AI, but there were limitations to practical use due to problems related to data, such as the limit of labeled data utilization and class imbalance. Therefore, this study aims to develop an AI model that can categorize the contract terms based on the FIDIC Yellow 2017(Federation Internationale Des Ingenieurs-Conseils Contract terms) standard in detail, rather than defining and classifying risk terms like previous research. A multi-text classification function is necessary because the contract terms that need to be reviewed in detail may vary depending on the scale and type of the project. To enhance the performance of the multi-text classification model, we developed the ELECTRA PLM (Pre-trained Language Model) capable of efficiently learning the context of text data from the pre-training stage, and conducted a four-step experiment to validate the performance of the model. As a result, the ensemble version of the self-developed ITB-ELECTRA model and Legal-BERT achieved the best performance with a weighted average F1-Score of 76% in the classification of 57 contract terms.

• Journal of the Korea Convergence Society
• /
• v.12 no.1
• /
• pp.57-63
• /
• 2021

Spoken language understanding, which aims to understand utterance as naturally as human would, are mostly focused on English language. In this paper, we construct a Korean language dataset for spoken language understanding, which is based on a conversational corpus between reservation system and its user. The domain of conversation is limited to restaurant reservation. There are 7 types of slot tags and 5 types of intent tags in 6857 sentences. When a model proposed in English-based research is trained with our dataset, intent classification accuracy decreased a little, while slot filling F1 score decreased significantly.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.4
• /
• pp.1275-1292
• /
• 2021

As a fine-grained classification problem, aspect-level sentiment classification predicts the sentiment polarity for different aspects in context. To address this issue, researchers have widely used attention mechanisms to abstract the relationship between context and aspects. Still, it is difficult to effectively obtain a more profound semantic representation, and the strong correlation between local context features and the aspect-based sentiment is rarely considered. In this paper, a hybrid attention capsule network for aspect-level sentiment classification (ABASCap) was proposed. In this model, the multi-head self-attention was improved, and a context mask mechanism based on adjustable context window was proposed, so as to effectively obtain the internal association between aspects and context. Moreover, the dynamic routing algorithm and activation function in capsule network were optimized to meet the task requirements. Finally, sufficient experiments were conducted on three benchmark datasets in different domains. Compared with other baseline models, ABASCap achieved better classification results, and outperformed the state-of-the-art methods in this task after incorporating pre-training BERT.