• Annual Conference on Human and Language Technology
• /
• 2016.10a
• /
• pp.300-304
• /
• 2016

본 논문에서는 한국어 의미역 결정을 순차열 분류 문제(Sequence Labeling Problem)가 아닌 순차열 변환 문제(Sequence-to-Sequence Learning)로 접근하였고, 구문 분석 단계와 자질 설계가 필요 없는 End-to-end 방식으로 연구를 진행하였다. 음절 단위의 RNN Search 모델을 사용하여 음절 단위로 입력된 문장을 의미역이 달린 어절들로 변환하였다. 또한 순차열 변환 문제의 성능을 높이기 위해 연구된 인풋-피딩(Input-feeding) 기술과 카피넷(CopyNet) 기술을 한국어 의미역 결정에 적용하였다. 실험 결과, Korean PropBank 데이터에서 79.42%의 레이블 단위 f1-score, 71.58%의 어절 단위 f1-score를 보였다.

• Journal of KIISE:Databases
• /
• v.35 no.2
• /
• pp.112-124
• /
• 2008

RDF and OWL are the primary base technologies for implementing Semantic Web. Recently, many researches related with them, or applying them into the other application domains, have been introduced. However, relatively little work has been done for securing the RDF and OWL data. In this article, we briefly introduce an RDF triple based model for specifying RDF access authorization related with RDF security. Next, to efficiently find the authorization conflict by RDF inference, we introduce a method using prime number graph labeling in detail. The problem of authorization conflict by RDF inference is that although the lower concept is permitted to be accessed, it can be inaccessible due to the disapproval for the upper concept. Because by the RDF inference, the lower concept can be interpreted into the upper concept. Some experimental results show that the proposed method using the prime number graph labeling has better performance than the existing simple method for the detection of the authorization conflict.

• Journal of the Korean Society of Manufacturing Process Engineers
• /
• v.19 no.7
• /
• pp.7-15
• /
• 2020

Recently, the fourth industrial revolution has been presented as a new paradigm and additive manufacturing (AM) has become one of the most important topics. For this reason, process monitoring for each cross-sectional layer of additive metal manufacturing is important. Particularly, deep learning can train a machine to analyze, optimize, and repair defects. In this paper, image classification is proposed by learning images of defects in the metal cross sections using the convolution neural network (CNN) image labeling algorithm. Defects were classified into three categories: crack, porosity, and hole. To overcome a lack-of-data problem, the amount of learning data was augmented using a data augmentation algorithm. This augmentation algorithm can transform an image to 180 images, increasing the learning accuracy. The number of training and validation images was 25,920 (80 %) and 6,480 (20 %), respectively. An optimized case with a combination of fully connected layers, an optimizer, and a loss function, showed that the model accuracy was 99.7 % and had a success rate of 97.8 % for 180 test images. In conclusion, image labeling was successfully performed and it is expected to be applied to automated AM process inspection and repair systems in the future.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.10
• /
• pp.1355-1360
• /
• 2020

In a map, placing the labels, corresponding to names or explanations of specific features, is called map labeling. In this paper, n points on a line are given, and placing rectangular labels for the points is considered. Particularly, the labels have a same height and their lower sides lie on a straight line in the upper of the line on which the given points are. The points and the labels are connected by polygonal lines, which are called leader lines. The leader lines are classified into straight leader lines and bended leader lines, where the straight leader line consists of only the vertical line and the bended leader line consists of vertical, horizontal, vertical lines. The problem is placing the labels to minimize the number of bended leader lines, and we propose an O(nlogn)-time algorithm, which improves the O(n2)-time algorithm previously provided in [13].

• The Korean Journal of Food And Nutrition
• /
• v.35 no.6
• /
• pp.537-542
• /
• 2022

This study was conducted to provide basic data on the amount of sodium and the setting of permissible error range of actual measurement, which is a problem for cabbage kimchi nutrients subject to labeling. The sample targeted was baechu (Kimchi cabbage) kimchi, which might have a large variation in sodium content by part of kimchi. Kimchi samples were collected twice from eight companies by season (spring, summer, fall, and winter). The average sodium content in kimchi samples was 619±87 mg/100 g (range, 534±63 mg/100 g to 783±40 mg/100 g). The error in average annual sodium content of abandonment kimchi (maximum value difference compared to the minimum value) was 26.8 to 64.3%. Sodium contents in kimchi produced in spring and summer were relatively low. However, deviation between individuals was large. It was found that cases exceeding the permissible error (120%) standard varied depending on the criteria for setting the amount of sodium. In addition, due to seasonal differences, sodium content in kimchi exceeded 120% of the labeling value. Thus, it is necessary to set standards suitable for characteristics of kimchi to prevent unintentional violations of labeling standards by raw materials and manufacturing processes.

• Journal of Korean Society of Transportation
• /
• v.13 no.3
• /
• pp.53-65
• /
• 1995

Proposed are model and its solution algorithm for determining the minimum number of high speed exits and their locations. While the previous researches on exit location aimed to minimize the average runway occuancy time (ROT) of an aircraft mix, the proposed approach is to find the minimum number of exits required to meet maximum allowable ROT. The rationale behind the approach is that the capacity of a runway increases as the ROT decreases down to some value, but not any more even though the ROT keep decreasing below the value. Hence, a maximum allowable ROT might be set up without declining the capacity. The problem is transformed into a shortest path problem on a specially constructed network and Dijkstra's labeling algorithms is employed to solve the problem A hypothetical example is provided to illustrate how the algorithms solves the problem.

• KIPS Transactions on Software and Data Engineering
• /
• v.5 no.11
• /
• pp.555-562
• /
• 2016

Semantic Role Labeling (SRL) is to determine the semantic relation of a predicate and its argu-ments in a sentence. But Korean semantic role labeling has faced on difficulty due to its different language structure compared to English, which makes it very hard to use appropriate approaches developed so far. That means that methods proposed so far could not show a satisfied perfor-mance, compared to English and Chinese. To complement these problems, we focus on suffix information analysis, such as josa (case suffix) and eomi (verbal ending) analysis. Korean lan-guage is one of the agglutinative languages, such as Japanese, which have well defined suffix structure in their words. The agglutinative languages could have free word order due to its de-veloped suffix structure. Also arguments with a single morpheme are then labeled with statistics. In addition, machine learning algorithms such as Support Vector Machine (SVM) and Condi-tional Random Fields (CRF) are used to model SRL problem on arguments that are not labeled at the suffix analysis phase. The proposed method is intended to reduce the range of argument instances to which machine learning approaches should be applied, resulting in uncertain and inaccurate role labeling. In experiments, we use 15,224 arguments and we are able to obtain approximately 83.24% f1-score, increased about 4.85% points compared to the state-of-the-art Korean SRL research.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.1
• /
• pp.24-29
• /
• 2022

Semantic role labeling task used in various natural language processing fields, such as information extraction and question answering systems, is the task of identifying the arugments for a given sentence and predicate. Predicate used as semantic role labeling input are extracted using lexical analysis results such as POS-tagging, but the problem is that predicate can't extract all linguistic patterns because predicate in korean language has various patterns, depending on the meaning of sentence. In this paper, we propose a korean predicate recognition method using neural network model with pre-trained embedding models and lexical features. The experiments compare the performance on the hyper parameters of models and with or without the use of embedding models and lexical features. As a result, we confirm that the performance of the proposed neural network model was 92.63%.

• Korean Journal of Computational Design and Engineering
• /
• v.16 no.4
• /
• pp.249-259
• /
• 2011

In this paper we present our work on the parameterized construction of virtual drivers' reach motion to seat belt, by using motion capture data. A user can generate a new reach motion by controlling a number of parameters. We approach the problem by using multiple sets of example reach motions and learning the relation between the labeling parameters and the motion data. The work is composed of three tasks. First, we construct a motion database using multiple sets of labeled motion clips obtained by using a motion capture device. This involves removing the redundancy of each motion clip by using PCA (Principal Component Analysis), and establishing temporal correspondence among different motion clips by automatic segmentation and piecewise time warping of each clip. Next, we compute motion blending functions by learning the relation between labeling parameters (age, hip base point (HBP), and height) and the motion parameters as represented by a set of PC coefficients. During runtime, on-line motion synthesis is accomplished by evaluating the motion blending function from the user-supplied control parameters.

• Journal of Computing Science and Engineering
• /
• v.3 no.3
• /
• pp.165-180
• /
• 2009

The HKIB, or Hankookilbo, test collections are two archives of Korean newswire stories manually categorized with semi-hierarchical or hierarchical category taxonomies. The base newswire stories were made available by the Hankook Ilbo (The Korea Daily) for research purposes. At first, Chungnam National University and KISTI collaborated to manually tag 40,075 news stories with categories by semi-hierarchical and balanced three-level classification scheme, where each news story has only one level-3 category (single-labeling). We refer to this original data set as HKIB-40075 test collection. And then Yonsei University and KISTI collaborated to select 20,000 newswire stories from the HKIB-40075 test collection, to rearrange the classification scheme to be fully hierarchical but unbalanced, and to assign one or more categories to each news story (multi-labeling). We refer to this modified data set as HKIB-20000 test collection. We benchmark a k-NN categorization algorithm both on HKIB-20000 and on HKIB-40075, illustrating properties of the collections, providing baseline results for future studies, and suggesting new directions for further research on Korean text categorization problem.