• Journal of the Korea Convergence Society
• /
• v.11 no.6
• /
• pp.23-28
• /
• 2020

A comment system is essential for communication on the Internet. However, there are also malicious comments such as inappropriate expression of others by exploiting anonymity online. In order to protect users from malicious comments, classification of malicious / normal comments is necessary, and this can be implemented as text classification. Text classification is one of the important topics in natural language processing, and studies using pre-trained models such as BERT and graph structures such as GCN and GAT have been actively conducted. In this study, we implemented a comment classification system using BERT, GCN, and GAT for actual published comments and compared the performance. In this study, the system using the graph-based model showed higher performance than the BERT.

• Journal of Korea Multimedia Society
• /
• v.19 no.1
• /
• pp.86-95
• /
• 2016

The main objective of HACCP(Hazard analysis and critical control points) system is to provide a systematic preventive approach how to control the risks in food production process. Practically, the diet classification process performed at the one of the beginning steps of the HACCP system, makes an important role of determining food safety risks and how to control them in every control point according to the different risk level of the diet. In this paper, we propose an automatic diet classification method for HACCP system using XML(eXtensible Markup Language). In order to guarantee the diet classification accuracy, we design the XML schema and attributes represents the relationship of every diet and ingredients analysing the HACCP diet classification rules. Based on the XML schema and document generation method, we develope the proposed system as client and server model that implemented XML based HACCP diet information generation module and integrated HACCP information management modules, respectively. Moreover, we show the efficiency of the proposed system with experiment results describing the school food diet information as XML documents and parsing the diet information.

• Journal of KIISE:Information Networking
• /
• v.36 no.3
• /
• pp.153-162
• /
• 2009

It is a challenging work for service operators to accurately classify different services, which runs on various wireless networks based upon numerous platforms. This works focuses on design and implementation of a classifier, which accurately classifies applications, which are captured horn WiBro Network. Notion of session is introduced for the classifier, instead of commonly used Flow to develop a classifier. Based on session information of given traffic, two classification algorithms are presented, Classification and Regression Tree and Support Vector Machine. Both algorithms are capable of classifying accurately and effectively with misclassification rate of 0.85%, and 0.94%, respectively. This work shows that classifier using CART provides ease of interpreting the result and implementation.

• Korean Journal of Remote Sensing
• /
• v.24 no.5
• /
• pp.507-515
• /
• 2008

Many studies have generally used a large number of pure pixels as an approach to training set design. The training set are used, however, varies between classifiers. In the recent research, it was reported that small mixed pixels between classes are actually more useful than larger pure pixels of each class in Support Vector Machine (SVM) classification. We evaluated a usability of small mixed pixels as a training set for the classification of high-resolution satellite imagery. We presented an advanced approach to obtain a mixed pixel readily, and evaluated the appropriateness with the land cover classification from IKONOS satellite imagery. The results showed that the accuracy of the classification based on small mixed pixels is nearly identical to the accuracy of the classification based on large pure pixels. However, it also showed a limitation that small mixed pixels used may provide insufficient information to separate the classes. Small mixed pixels of the class border region provide cost-effective training sets, but its use with other pixels must be considered in use of high-resolution satellite imagery or relatively complex land cover situations.

• Geomechanics and Engineering
• /
• v.37 no.6
• /
• pp.539-554
• /
• 2024

This study explores development of prediction model for seismic site classification through the integration of machine learning techniques with horizontal-to-vertical spectral ratio (HVSR) methodologies. To improve model accuracy, the research employs outlier detection methods and, synthetic minority over-sampling technique (SMOTE) for data balance, and evaluates using seven machine learning models using seismic data from KiK-net. Notably, light gradient boosting method (LGBM), gradient boosting, and decision tree models exhibit improved performance when coupled with SMOTE, while Multiple linear regression (MLR) and Support vector machine (SVM) models show reduced efficacy. Outlier detection techniques significantly enhance accuracy, particularly for LGBM, gradient boosting, and voting boosting. The ensemble of LGBM with the isolation forest and SMOTE achieves the highest accuracy of 0.91, with LGBM and local outlier factor yielding the highest F1-score of 0.79. Consistently outperforming other models, LGBM proves most efficient for seismic site classification when supported by appropriate preprocessing procedures. These findings show the significance of outlier detection and data balancing for precise seismic soil classification prediction, offering insights and highlighting the potential of machine learning in optimizing site classification accuracy.

• Journal of the Earthquake Engineering Society of Korea
• /
• v.20 no.4
• /
• pp.245-256
• /
• 2016

In the companion paper (I - Database and Site Response Analyses), site-specific response analyses were performed at more than 300 domestic sites. In this study, a new site classification system and design response spectra are proposed using results of the site-specific response analyses. Depth to bedrock (H) and average shear wave velocity of soil above the bedrock ($V_{S,Soil}$) were adopted as parameters to classify the sites into sub-categories because these two factors mostly affect site amplification, especially for shallow bedrock region. The 20 m of depth to bedrock was selected as the initial parameter for site classification based on the trend of site coefficients obtained from the site-specific response analyses. The sites having less than 20 m of depth to bedrock (H1 sites) are sub-divided into two site classes using 260 m/s of $V_{S,Soil}$ while the sites having greater than 20 m of depth to bedrock (H2 sites) are sub-divided into two site classes at $V_{S,Soil}$ equal to 180 m/s. The integration interval of 0.4 ~ 1.5 sec period range was adopted to calculate the long-period site coefficients ($F_v$) for reflecting the amplification characteristics of Korean geological condition. In addition, the frequency distribution of depth to bedrock reported for Korean sites was also considered in calculating the site coefficients for H2 sites to incorporate sites having greater than 30 m of depth to bedrock. The relationships between the site coefficients and rock shaking intensity were proposed and then subsequently compared with the site coefficients of similar site classes suggested in other codes.

• ETRI Journal
• /
• v.22 no.2
• /
• pp.1-10
• /
• 2000

Pattern recognition is one of the most common problems encountered in engineering and scientific disciplines, which involves developing prediction or classification models from historic data or training samples. This paper introduces a new approach, called the Representational Capability (RC) algorithm, to handle pattern recognition problems using radial basis function (RBF) models. The RC algorithm has been developed based on the mathematical properties of the interpolation and design matrices of RBF models. The model development process based on this algorithm not only yields the best model in the sense of balancing its parsimony and generalization ability, but also provides insights into the design process by employing a design parameter (${\delta}$). We discuss the RC algorithm and its use at length via an illustrative example. In addition, RBF classification models are developed for heart disease diagnosis.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.33B no.6
• /
• pp.102-110
• /
• 1996

A new design scheme of a binary decision tree is proposed. In this scheme a binary decision tree is constructed by using genetic algorithm and FCM algorithm. At each node optimal or near-optimal feature subset is selected which optimizes fitness function in genetic algorithm. The fitness function is inversely proportional to classification error, balance between cluster, number of feature used. The binary strings in genetic algorithm determine the feature subset and classification results - error, balance - form fuzzy partition matrix affect reproduction of next genratin. The proposed design scheme is applied to the tire tread patterns and handwriteen alphabetic characters. Experimental results show the usefulness of the proposed scheme.

• Archives of design research
• /
• v.16 no.3
• /
• pp.81-90
• /
• 2003

The purpose of the study is to propose a new way of classification for products and to forecast the future of products through the physical factor and the mental factor as human natures. For the purpose of the study, the research was carried out in three ways. Firstly, the study considered the evolutional process of products through human natures. At this stage, the study defined that the physical ability and the mental ability of human are the cores of the product's evolution. Secondly, for understanding human evolution, the study set up two types of future humans . Finally, the study classified products by the physical factor and the mental factor as human natures with the aspect of embryology. As the results, the study illustrated two different species of products and their futures.

• The Journal of the Korea Contents Association
• /
• v.10 no.5
• /
• pp.99-106
• /
• 2010

Multilayer perceptrons(MLPs) or feed-forward neural networks are widely applied to many areas based on their function approximation capabilities. When implementing MLPs for application problems, we should determine various parameters and training methods. In this paper, we discuss the design of MLPs especially for pattern classification problems. This discussion includes how to decide the number of nodes in each layer, how to initialize the weights of MLPs, how to train MLPs among various error functions, the imbalanced data problems, and deep architecture.