• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2021년도 춘계학술발표대회
• /
• pp.408-410
• /
• 2021

In the NLP field, the pre-training model BERT launched by the Google team in 2018 has shown amazing results in various tasks in the NLP field. Subsequently, many variant models have been derived based on the original BERT, such as RoBERTa, ERNIEBERT and so on. In this paper, the WWMBERT (Whole Word Masking BERT) model suitable for Chinese text tasks was used as the baseline model of our experiment. The experiment is mainly for "Text-level Chinese text classification tasks" are improved, which mainly combines Tapt (Task-Adaptive Pretraining) and "Multi-Sample Dropout method" to improve the model, and compare the experimental results, experimental data sets and model scoring standards Both are consistent with the official WWMBERT model using Accuracy as the scoring standard. The official WWMBERT model uses the maximum and average values of multiple experimental results as the experimental scores. The development set was 97.70% (97.50%) on the "text-level Chinese text classification task". and 97.70% (97.50%) of the test set. After comparing the results of the experiments in this paper, the development set increased by 0.35% (0.5%) and the test set increased by 0.31% (0.48%). The original baseline model has been significantly improved.

• Journal of Trauma and Injury
• /
• 제25권4호
• /
• pp.139-144
• /
• 2012

Purpose: Hypovolemic shock is the leading cause of death in multiple trauma patients with pelvic bone fracures. The purpose of this study was to develop a simple prediction method for injury severity and amount of hemorrhage via an analysis of the correlation between the site of pelvic bone fracture and the amount of transfusion and to verify the usefulness of the such a simple scoring system. Methods: We analyzed retrospectively the medical records and radiologic examination of 102 patients who had been diagnosed as having a pelvic bone fracture and who had visited the Emergency Department between January 2007 and December 2011. Fracture sites in the pelvis were confirmed and re-classified anatomically as pubis, ilium or sacrum. A multiple linear regression analysis was performed on the amount of transfusion, and a simplified scoring system was developed. The predictive value of the amount of transfusion for the scoring system as verified by using the receiver operating characteristics (ROC). The area under the curve of the ROC was compared with the injury severity score (ISS). Results: From among the 102 patients, 97 patients (M:F=68:29, mean $age=46.7{\pm}16.6years$) were enrolled for analysis. The average ISS of the patients was $16.2{\pm}7.9$, and the average amount of packed RBC transfusion for 24 hr was $3.9{\pm}4.6units$. The regression equation resulting from the multiple linear regression analysis was 'packed RBC units=1.40${\times}$(sacrum fracture)+1.72${\times}$(pubis fracture)+1.67${\times}$(ilium fracture)+0.36' and was found to be suitable (p=0.005). We simplified the regression equation to 'Pelvic Bleeding Score=sacrum+pubis+ilium.' Each fractured site was scored as 0(no fracture) point, 1(right or left) point, or 2(both) points. Sacrum had only 0 or 1 point. The score ranged from 0 to 5. The area under the curve (AUC) of the ROC was 0.718 (95% CI: 0.588-0.848, p=0.009). For an upper Pelvis Bleeding Score of 3 points, the sensitivity of the prediction for a massive transfusion was 71.4%, and the specificity was 69.9%. Conclusion: We developed a simplified scoring system for the anatomical fracture sites in the pelvis to predict the requirement for a transfusion (Pelvis Bleeding Score (PBS)). The PBS, compared with the ISS, is considered a useful predictor of the need for a transfusion during initial management.

• 정보보호학회논문지
• /
• 제17권2호
• /
• pp.29-40
• /
• 2007

입력이 되는 지문 영상의 품질은 지문 인식 시스템의 전체 인식 성능에 다양하게 영향을 미친다. 지문 영상의 품질은 추출, 정합 등의 알고리즘 내에서도 중요한 정보로 활용되는데, 특히 센서 입력부에서의 품질 측정 및 평가는 불필요한 처리 과정을 줄여 시스템의 전반적인 안정성을 유도하는데 사용되는 만큼 매우 중요한 과정이 된다. 본 논문에서는 기존의 많은 연구 결과들과는 달리 센서, 인식 알고리즘, 손가락의 특성에 독립적인 지문 영상 품질을 정의하였다. 또한 이의 측정을 위하여 지문 영상의 지역적 특성들만을 이용한 새로운 품질 측정 방법도 제안하였다. 제안하는 품질 측정 방법은 블록별 품질 특성 분류 결과를 이용하여 전체 지문 영상의 품질을 수치화하는 것으로, 본 논문에서는 FVC에서 사용된 몇 가지 서브 데이터베이스들을 이용하여 제안하는 방법의 유용함을 증명하였다. 실험 결과는 NIST의 NFIQ, AWARE사의 QualityCheck 품질 측정 알고리즘들과 비교되었는데, 이들 알고리즘들에 비해 제안하는 측정 방법이 인간의 시각적 분류 기준과 유사한 결과를 보였다.

• 한국컴퓨터정보학회논문지
• /
• 제25권12호
• /
• pp.145-153
• /
• 2020

최근 소셜 빅데이터를 대상으로 한 키워드 분석은 고객 관점의 의견이나 불만 사항을 추출하기 위한 목적으로 광범위하게 활용되고 있다. 이와 관련하여, 이전 연구에서는 키워드 분석의 정확도를 높이기 위해 응집도 점수를 활용한 방법을 제안하였으나, 리뷰의 수가 적을 경우 오류율이 증가하는 문제가 있었다. 본 논문에서는 응집도 점수 기반 알고리즘으로부터 추출된 키워드에 대해 간소화된 형태소 분석 단계를 후처리 형태로 적용함으로써 키워드 추출의 정확도를 개선하고자 하였다. 제안 방법은 입력 데이터가 주어질 때마다 필요한 형태소 분석 규칙을 점증적으로 추가할 수 있도록 지원함으로써, 사전의 크기를 최소화하고 분석의 효율을 높이고자 하였다. 또한 대화형 규칙 입력 시스템을 제공하여 분석 규칙 추가에 드는 노력을 최소화하고자 하였다. 제안 방법을 검증하기 위해 온라인에서 수집된 실제 리뷰를 대상으로 실험을 수행하였으며, 제안 방법을 적용할 경우 오류율이 기존 10%에서 1%로 개선되는 동시에, 5,000개의 리뷰 처리에 450ms가 소요되어 실시간 처리가 가능한 수준임을 확인하였다.

• 한국경영과학회지
• /
• 제8권2호
• /
• pp.67-71
• /
• 1983

The purpose of this paper is to propose a method of constructing a school timetable under consideration of subjects, teachers, classes, rooms, activity types and so on. The method gives an acceptable timetable which satisfies all dispensable requirements and different timetables also are obtained by changing the order of assignments. The resulting timetables are evaluated and compared by scoring scheme which can be modified by users. The results achieved using real data have been encouraging.

• 수산해양기술연구
• /
• 제49권4호
• /
• pp.469-482
• /
• 2013

This study identified problems of the existing ecosystem-based fisheries assessment approach, and suggested new methods for scoring risk and for the estimation of fishery risk index. First, risk scores of zero to two for target and limit reference points for each indicator were replaced by those of zero to three, and the risk scores were calculated from new formulae which were developed in this study. Second, a new method for estimating fishery risk index (FRI) was developed in this study, considering the level of indicators. New method was applied to the Korean large purse seine fishery, large pair trawl fishery and drag net fishery. More precise and detailed risk scores were obtained from the new method, which can explain the risks by the wider range of both risk levels for 'better than target' and 'beyond limit'. The new method for estimating FRI could avoid the basic problem related with duplicated computations of fishery-level indicators, which improved the estimated FRI to be more accurate. Also, a method for estimating variance of FRI using the bootstrap was proposed in this study.

• 대한임상독성학회지
• /
• 제12권1호
• /
• pp.14-21
• /
• 2014

Purpose: The objective of this study was to develop a new scoring tool that is comprehensively applicable and predicts fatality within 24 h of intoxication. Methods: This was a cohort study conducted in two emergency medical centers from 2011 to 2012. We identified factors associated with severe/fatality. Through a discriminant analysis, we devised the aBIG (age, Base deficit, Infection, and Glasgow coma scale) score. To compare the ability of aBIG to predict intoxication severity with that of previous scoring systems such as APACHE II, MODS, SAPS IIe, and SOFA, we determined the receiver operating characteristic curves of each variable in predicting severe-to-fatal toxicity. Results: Compared with the mild/moderate toxicity group (n=211), the severe/fatal group (n=143) had higher incidences of metabolic acidosis, infection, serious mental change, QTc prolongation and hepato-renal failure. Age, base deficit, infection-WBC count, and Glasgow Coma Scale were independently associated with severe/fatal poisoning. These variables were combined into the poisoning "aBIG" score [$0.28{\times}$Age group+$0.38{\times}WBC$ count/$10^3+0.52{\times}$Base deficit+$0.64{\times}$(15-GCS)], which were each calculated to have an area under the curve of 0.904 (95% confidence interval: 0.868-0.933). The aBIG poisoning score had an equivalent level of severity predictability as APACHE II and a superior than MODS, SOFA, and SAPS IIe. Conclusion: We developed a simplified scoring system using the four variables of age, base deficit, infected leukocytosis, and GCS. The poisoning aBIG score was a simple method that could be performed rapidly on admission to evaluate severity of illness and predict fatal severity in patients with acute intoxications.

• 한국과학교육학회지
• /
• 제40권3호
• /
• pp.321-336
• /
• 2020

본 연구는 실제 교실에서 이루어진 학생의 과학 논증과정을 기계학습을 활용한 자동 채점에 적용함으로써, 논증 자동 채점의 가능성 및 개선 방향을 탐색한다. 분자 구조에 대한 고등학생의 과학 논증수업 중 발생한 2,605개의 모든 발화를 대상으로 연구를 진행하였다. 지도 학습을 위해 5가지의 논증 요소로 발화를 분류하였고, 분류된 발화를 대상으로 텍스트 전처리를 수행하였다. 전처리된 학생 발화를 활용하여 서포트 벡터 머신, 의사결정나무, 랜덤 포레스트, 인공신경망의 기계 학습 방법으로 자동 채점 모델을 구성하였다. 불용어 처리가 되지 않은 학생 발화를 활용한 자동 채점의 결과 랜덤 포레스트의 정확도는 65.96%, kappa는 0.5298의 유미한 결과를 얻었다. 불용어 처리를 수행한 학생 발화를 활용한 새로운 채점 모델의 결과 채점의 정확도가 크게 변화하지 않음에도 논증 발화 중 과학 용어 및 논증 요소의 담화표지가 채점 모델의 분류 기준이 되는 결과를 얻었다. 또한 인간 전문가의 논증 채점 과정을 분석하여 얻어진 전문가 형태소를 자동 채점 모델에 생성 규칙 알고리즘으로 적용하였다. 그 결과 의사결정나무에서 반박에 대한 재현율(recall)이 21.74% 증가하였다. 이에 본 연구 결과는 과학 교육 연구에서 기계 학습 및 논증에 대한 자동 채점의 활용 가능성과 연구 방향성을 제안하였다.

• 한국토양비료학회지
• /
• 제45권6호
• /
• pp.1237-1241
• /
• 2012

Heavy metal pollution in agricultural field has been a critical issue in worldwide. For this reason, remediation technologies for heavy metal polluted soil are applied especially near at the abandoned metal mine. Soil quality analysis is also an important factor for proper management in heavy metal polluted agricultural field. In this study, scoring function was utilized to evaluate soil quality in heavy metal polluted agricultural field. Among other soil properties, bulk density, soil pH, EC, $NH_4$-N, $NO_3$-N, and cation exchange capacity (CEC) were determined for minimum data set (MDS) with principal component analysis. Result showed that both upland and paddy soil contaminated with heavy metal were not suitable for crop growth except scoring of soil pH for paddy soil and CEC for upland soil. This result might indicate that chemical stabilization technology with chemical amendment could be adapted for remediation method for heavy metal polluted agiclutural field not only for heavy metal immobilization but also enhancement of soil condition for crop growth.

• 한국전문물리치료학회지
• /
• 제19권4호
• /
• pp.46-54
• /
• 2012

Ely's test is commonly used to assess rectus femoris muscle flexibility. however, a reliability limit of this test was demonstrated by a previous study. In this study, we present an alternative method using an application for a digital horizontal level in a smartphone (DHLS) to complement the reliability limit of Ely's test. The aim of this study was to examine the reliability of Ely's test using DHLS on the pelvis, compared to using visual observation (VO) of pelvic and hip motions. Nineteen patients with lumbar extension-rotation syndrome were recruited for this study. An examiner examined the rectus femoris flexibility (both pass/fail and goniometer scoring) through Ely's test using both DHLS and VO. A retest session was completed two hours later for within-day reliability and seven days later for between-day intra-rater test-retest reliability. Results showed higher Kappa values for pass/fail scoring and higher intraclass correlation coefficient values for goniometer scoring in Ely's test using DHLS, compared to using VO. Measurement error and Bland and Altman plots further demonstrated the degree of intra-rater variance during Ely's test using DHLS in a clinical setting, compared to using VO. These results demonstrated that Ely's test using DHLS showed acceptable reliability compared to using VO. Ely's test using DHLS could be widely used for measuring the rectus femoris muscle flexibility in patients with lumbar extension-rotation syndrome, although the inter-rater reliability needs to be established first.