• Journal of Intelligence and Information Systems
• /
• v.29 no.3
• /
• pp.419-437
• /
• 2023

As the number of thin filers in Korea surpasses 12 million, there is a growing interest in enhancing the accuracy of assessing their credit default risk to generate additional revenue. Specifically, researchers are actively pursuing the development of default prediction models using machine learning and deep learning algorithms, in contrast to traditional statistical default prediction methods, which struggle to capture nonlinearity. Among these efforts, Graph Neural Network (GNN) architecture is noteworthy for predicting default in situations with limited data on thin filers. This is due to their ability to incorporate network information between borrowers alongside conventional credit-related data. However, prior research employing graph neural networks has faced limitations in effectively handling diverse categorical variables present in credit information. In this study, we introduce the Transformer embedded Graph Convolutional Network (TeGCN), which aims to address these limitations and enable effective default prediction for thin filers. TeGCN combines the TabTransformer, capable of extracting contextual information from categorical variables, with the Graph Convolutional Network, which captures network information between borrowers. Our TeGCN model surpasses the baseline model's performance across both the general borrower dataset and the thin filer dataset. Specially, our model performs outstanding results in thin filer default prediction. This study achieves high default prediction accuracy by a model structure tailored to characteristics of credit information containing numerous categorical variables, especially in the context of thin filers with limited data. Our study can contribute to resolving the financial exclusion issues faced by thin filers and facilitate additional revenue within the financial industry.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• v.34S no.11
• /
• pp.138-147
• /
• 1997

The ultrasonic images are corrupted by the granular pattern noise - a speckle noise. The speckle exist in the type of coherent imaging systems, and the speckle is the signal independent and multiplicative noise. In this paepr, we derive two filters using the gradient and symmetry. One is a noise suppression filter which removes noise while preserves the edges. It is named the ASRF-GS (Adaptive Speckle Removal Filer - Gradient and Symmetry). And the other is a edge detection filter which obtains the thin edge map, called the EDUGS(Edge Detection Using Gradient and Symmetry). The performance of the proposed noise suppression filter is evaluated by the IMPV(SNR improvement) and the Speckle Index(SI), and the perforamnce of the edge detection is evaluated by the edge detection error rate. According to the evaluated method, The SI reduced about 0.035, The IMPV improved about 1.265(dB), and the edge detection error rate is about 17.5%.

• Proceedings of the Korean Institute of Electrical and Electronic Material Engineers Conference
• /
• 2010.06a
• /
• pp.383-383
• /
• 2010

TiN is one of the mostly used barrier materials in copper metallization because of low friction coefficient and superior electrical properties. We need to investigate for the etching characteristic of TiN. In this study, we investigated about etching characteristic of TiN using $BCl_3$/Ar inductively coupled plasma system. The etch rate was measured by a depth pro filer. The chemical etching reactions of the TiN surface was investigated X-ray photoelectron spectroscopy.

• Journal of Korean Society of Occupational and Environmental Hygiene
• /
• v.29 no.3
• /
• pp.310-321
• /
• 2019

Objectives: The purpose of this study was to investigate types and characteristics of chemical substances used in LCD(Liquid crystal display) panel manufacturing process. Methods: The LCD panel manufacturing process is divided into the fabrication(fab) process and module process. The use of chemical substances by process was investigated at four fab processes and two module processes at two domestic TFT-LCD(Thin film transistor-Liquid crystal display) panel manufacturing sites. Results: LCD panels are manufactured through various unit processes such as sputtering, chemical vapor deposition(CVD), etching, and photolithography, and a range of chemicals are used in each process. Metal target materials including copper, aluminum, and indium tin oxide are used in the sputtering process, and gaseous materials such as phosphine, silane, and chlorine are used in CVD and dry etching processes. Inorganic acids such as hydrofluoric acid, nitric acid and sulfuric acid are used in wet etching process, and photoresist and developer are used in photolithography process. Chemical substances for the alignment of liquid crystal, such as polyimides, liquid crystals, and sealants are used in a liquid crystal process. Adhesives and hardeners for adhesion of driver IC and printed circuit board(PCB) to the LCD panel are used in the module process. Conclusions: LCD panels are produced through dozens of unit processes using various types of chemical substances in clean room facilities. Hazardous substances such as organic solvents, reactive gases, irritants, and toxic substances are used in the manufacturing processes, but periodic workplace monitoring applies only to certain chemical substances by law. Therefore, efforts should be made to minimize worker exposure to chemical substances used in LCD panel manufacturing process.

• The Journal of the Convergence on Culture Technology
• /
• v.10 no.3
• /
• pp.467-474
• /
• 2024

This study analyses customer data from a US online lending platform to empirically document the discriminatory treatment that low- and middle-income borrowers face in financial markets. Researchers are using financial data from nearly 2.93 million loans between 2007~2020 of the Lending Club on the open-source Kaggle platform. We find that thin-filers borrowers, especially those with lower credit scores, receive loans at higher interest rates. This discriminatory treatment undermines financial inclusion and has the potential to increase social inequality. The significance of this research is that it sheds substantial light on the problem of inequality in financial markets and, based on the findings, suggests concrete measures to ensure equitable access to finance for all customers and enhance sustainable financial inclusion. In doing so, we propose a shift towards enhancing the social responsibility of institutions.

• Asia-Pacific Journal of Business Venturing and Entrepreneurship
• /
• v.15 no.6
• /
• pp.135-154
• /
• 2020

This paper analyzes the 5,521 samples of the small and medium-sized businesses(SMBs) obtained from the Korea Credit Guarantee Fund. From January 2014 to September 2019, 85% of the SMBs have 5 or fewer full-time employees. The proportion of SMBs is overwhelmed by the elderly men, and most founders are the CEO. Also, about 87% of the workplace types are rented, while 64% of the CEO's residence types are owner-occupation. 47% of the financial grade score is less than 10 points out of 100 and 80% of SMBs have less than 200 million won of the loan guarantee. In particular, the total guarantee loan amount or the days of net guarantee have significantly positive relations with the working period of the CEO in the same industry, the number of employees, the operation period of SMBs, and the corporate business type. In the case of the financial grading score which has the highest weight in overall credit rating gets higher with the higher number of employees, the longer the operation period, and the corporate business type. However, the quantified non-financial grading score has no significant relationship with other explanatory variables, except for the corporate business type. This implies that a non-financial grade score is measured by other determinants that are not observed by the Korea credit guarantee fund. The pure non-financial grade score has positive relations with the working period of the CEO. Overall, this paper would help Korean SMBs upgrade their credit ratings and expand the money supply when there is no standardized credit rating model or no publicly available evaluation criteria for SMBs. We expect this paper provides important insights for further research and policy-makers for SMBs. In particular, to address the financial needs of thin-filers such as SMBs, technology-based financial services (TechFin) would use alternative data to evaluate the financial capabilities of thin-filers and to develop new financial services.

• Asia-Pacific Journal of Business Venturing and Entrepreneurship
• /
• v.18 no.5
• /
• pp.77-90
• /
• 2023

This study aims to investigate the potential of alternative credit assessment through Social Networking Sites (SNS) as a complementary tool to conventional loan review processes. It seeks to discern the impact of SNS usage characteristics and loan product attributes on credit loan repayment. To achieve this objective, we conducted a binomial logistic regression analysis examining the influence of SNS usage patterns, loan characteristics, and personal attributes on credit loan conditions, utilizing data from Company A's credit loan program, which integrates SNS data into its actual loan review processes. Our findings reveal several noteworthy insights. Firstly, with respect to profile photos that reflect users' personalities and individual characteristics, individuals who choose to upload photos directly connected to their personal lives, such as images of themselves, their private circles (e.g., family and friends), and photos depicting social activities like hobbies, which tend to be favored by individuals with extroverted tendencies, as well as character and humor-themed photos, which are typically favored by individuals with conscientious traits, demonstrate a higher propensity for diligently repaying credit loans. Conversely, the utilization of photos like landscapes or images concealing one's identity did not exhibit a statistically significant causal relationship with loan repayment. Furthermore, a positive correlation was observed between the extent of SNS usage and the likelihood of loan repayment. However, the level of SNS interaction did not exert a significant effect on the probability of loan repayment. This observation may be attributed to the passive nature of the interaction variable, which primarily involves expressing sympathy for other users' comments rather than generating original content. The study also unveiled the statistical significance of loan duration and the number of loans, representing key characteristics of loan portfolios, in influencing credit loan repayment. This underscores the importance of considering loan duration and the quantity of loans as crucial determinants in the design of microcredit products. Among the personal characteristic variables examined, only gender emerged as a significant factor. This implies that the loan program scrutinized in this analysis does not exhibit substantial discrimination based on age and credit scores, as its customer base predominantly consists of individuals in their twenties and thirties with low credit scores, who encounter challenges in securing loans from traditional financial institutions. This research stands out from prior studies by empirically exploring the relationship between SNS usage and credit loan repayment while incorporating variables not typically addressed in existing credit rating research, such as profile pictures. It underscores the significance of harnessing subjective, unstructured information from SNS for loan screening, offering the potential to mitigate the financial disadvantages faced by borrowers with low credit scores or those ensnared in short-term liquidity constraints due to limited credit history a group often referred to as "thin filers." By utilizing such information, these individuals can potentially reduce their credit costs, whereas they are supposed to accrue a more substantial financial history through credit transactions under conventional credit assessment system.