• Annual Conference on Human and Language Technology
• /
• 2009.10a
• /
• pp.213-218
• /
• 2009

본 논문에서는 무제한 텍스트 입력이 가능한 파서에서 오류분석을 통한 성능 향상을 이루고자 한다. 우선 코퍼스로부터 자동학습에 의해서 구문 분석 모델을 만들고 이를 평가하여 발생하는 오류를 분석한다. 오류를 감소시킬 수 있는 언어 특성이 반영된 자질을 추가하여 성능을 향상시키고자 한다. 세종 코퍼스를 10-fold cross validation으로 평가할 때, 한국어의 특성을 반영한 자질 추가로 1%이상의 성능 향상을 이루었다.

• The Korean Journal of Applied Statistics
• /
• v.29 no.7
• /
• pp.1257-1270
• /
• 2016

This study deals with the prediction of the total number of movie audiences as a measure for the box office. Prediction is performed by classification techniques of data mining such as decision tree, multilayer perceptron(MLP) neural network model, multinomial logit model, and support vector machine over time such as before movie release, release day, after release one week, and after release two weeks. Predictors used are: online word-of-mouth(OWOM) variables such as the portal movie rating, the number of the portal movie rater, and blog; in addition, other variables include showing the inherent properties of the film (such as nationality, grade, release month, release season, directors, actors, distributors, the number of audiences, and screens). When using 10-fold cross validation technique, the accuracy of the neural network model showed more than 90 % higher predictability before movie release. In addition, it can be seen that the accuracy of the prediction increases by adding estimates of the final OWOM variables as predictors.

• Journal of the Korea Society of Computer and Information
• /
• v.19 no.1
• /
• pp.111-118
• /
• 2014

Current friend recommendation system on Twitter primarily recommends the most influential twitter. However, this way of recommendation has drawbacks where it does not recommend the users of which attributes of interests are similar to theirs. Since users want other users of which attributes are similar, this study implements follower recommendation system based on the similarity of twitter node informations. The data in this study is from SNAP(Stanford Network Analysis Platform), and it consists of twitter node information of which number of followers is over 10,000 and twitter link information. We used the SNAP data as a training data, and generated a classifier which recommends and predicts the relation between followers. We evaluated the classifier by 10-Fold Cross validation. Once two twitter node informations are given, our model can recommend the relationship of the two twitters as one of following such as: FoFo(Follower Follower), FoFr(Follower Friend), NC(Not Connected).

• KIPS Transactions on Software and Data Engineering
• /
• v.3 no.10
• /
• pp.421-428
• /
• 2014

Like Alzheimer's disease, Parkinson's Disease(PD) is one of the most common neurodegenerative brain disorders. PD results from the deterioration of dopaminergic neurons in the brain region called the substantia nigra. Currently there is no cure for PD, but diagnosing in its early stage is important to provide treatments for relieving the symptoms and maintaining quality of life. Unlike many diagnosis methods of PD which use a single biomarker, we developed a diagnosis method that uses both biochemical biomarkers and imaging biomarkers. Our method uses ${\alpha}$-synuclein protein levels in the cerebrospinal fluid and diffusion tensor images(DTI). It achieved an accuracy over 91.3% in the 10-fold cross validation, and the best accuracy of 72% in an independent testing, which suggests a possibility for early detection of PD. We also analyzed the characteristics of the brain fiber pathways of Parkinson's disease patients and normal elderly people.

• Journal of Digital Contents Society
• /
• v.14 no.4
• /
• pp.581-587
• /
• 2013

In this paper, we propose a method of sentiment classification which uses Levenshtein distance. We generate BOW(Bag-Of-Word) applying Levenshtein daistance in sentiment features and used it as the training set. Then the machine learning algorithms we used were SVMs(Support Vector Machines) and NB(Naive Bayes). As the data set, we gather 2,385 reviews of movies from an online movie community (Daum movie service). From the collected reviews, we pick sentiment words up manually and sorted 778 words. In the experiment, we perform the machine learning using previously generated BOW which was applied Levenshtein distance in sentiment words and then we evaluate the performance of classifier by a method, 10-fold-cross validation. As the result of evaluation, we got 85.46% using Multinomial Naive Bayes as the accuracy when the Levenshtein distance was 3. According to the result of the experiment, we proved that it is less affected to performance of the classification in spelling errors in documents.

• The Journal of Korean Association of Computer Education
• /
• v.7 no.4
• /
• pp.51-60
• /
• 2004

This paper proposes an automated coding system of Korean standard industry/occupation for census which reduces a lot of cost and labor for manual coding. The proposed system converts natural language responses on survey questionnaires into corresponding numeric codes using information retrieval techniques and document classification algorithm. The system was experimented with 46,762 industry records and occupation 36,286 records using 10-fold cross -validation evaluation method. As experimental results, the system show 87.08% and 66.08% production rates when classifying industry records into level 2 and level 5 codes respectively. The system shows slightly lower performances on occupation code classification. We expect that the system is enough to be used as a semi-automate coding system which can minimize manual coding task or as a verification tool for manual coding results though it has much room to be improved as an automated coding system.

• The KIPS Transactions:PartD
• /
• v.19D no.1
• /
• pp.21-28
• /
• 2012

Predication of protein interaction sites for monomer structures can reduce the search space for protein docking and has been regarded as very significant for predicting unknown functions of proteins from their interacting proteins whose functions are known. In the other hand, the prediction of interaction sites has been limited in crystallizing weakly interacting complexes which are transient and do not form the complexes stable enough for obtaining experimental structures by crystallization or even NMR for the most important protein-protein interactions. This work reports the calculation of 3D surface patches of complex structures and their properties and a machine learning approach to build a predictive model for the 3D surface patches in interaction and non-interaction sites using support vector machine. To overcome classification problems for class imbalanced data, we employed an under-sampling technique. 9 properties of the patches were calculated from amino acid compositions and secondary structure elements. With 10 fold cross validation, the predictive model built from SVM achieved an accuracy of 92.7% for classification of 3D patches in interaction and non-interaction sites from 147 complexes.

• Journal of Korea Water Resources Association
• /
• v.48 no.6
• /
• pp.439-449
• /
• 2015

In this study, reliable future runoff projections based on RCPs for Yongdam dam watershed was performed using SWAT model, which was validated by k-fold cross validation method, and investigated the factors that cause the differences with respect to runoff projections between this study and previous studies. As a result, annual average runoff compared to baseline runoff would increase 17.7% and 26.1% in 2040s and 2080s respectively under RCP8.5 scenario, and 21.9% and 44.6% in 2040s and 2080s respectively under RCP4.5 scenario. Comparing the results to previous studies, minimum and maximum differences between runoff projections over different studies were 10.3% and 53.2%, even though runoff was projected by the same rainfall-runoff model. SWAT model has 27 parameters and physically based complex structure, so it tends to make different results by the model users' setting. In the future, it is necessary to reduce the cause of difference to generate standard runoff scenarios.

• Korean Journal of Remote Sensing
• /
• v.37 no.6_2
• /
• pp.1891-1900
• /
• 2021

Fine particulate matter (FPM; diameter ≤ 2.5 ㎛) is frequently found in metropolitan areas due to activities associated with rapid urbanization and population growth. Many adolescents spend a substantial amount of time at school where, for various reasons, FPM generated outdoors may flow into indoor areas. The aims of this study were to estimate FPM concentrations and categorize types of FPM in schools. Meteorological and chemical variables as well as satellite-based aerosol optical depth were analyzed as input data in a random forest model, which applied 10-fold cross validation and a grid-search method, to estimate school FPM concentrations, with four statistical indicators used to evaluate accuracy. Loose and strict standards were established to categorize types of FPM in schools. Under the former classification scheme, FPM in most schools was classified as type 2 or 3, whereas under strict standards, school FPM was mostly classified as type 3 or 4.

• The Journal of the Acoustical Society of Korea
• /
• v.27 no.3
• /
• pp.111-118
• /
• 2008

In this paper, we propose a new automatic taxonomy generation algorithm for the audio genre classification. The proposed algorithm automatically generates hierarchical taxonomy based on the estimated classification accuracy at all possible nodes. The estimation of classification accuracy in the proposed algorithm is conducted by applying the training data to classifier using k-fold cross validation. Subsequent classification accuracy is then to be tested at every node which consists of two clusters by applying one-versus-one support vector machine. In order to assess the performance of the proposed algorithm, we extracted various features which represent characteristics such as timbre, rhythm, pitch and so on. Then, we investigated classification performance using the proposed algorithm and previous flat classifiers. The classification accuracy reaches to 89 percent with proposed scheme, which is 5 to 25 percent higher than the previous flat classification methods. Using low-dimensional feature vectors, in particular, it is 10 to 25 percent higher than previous algorithms for classification experiments.