• Journal of Intelligence and Information Systems
• /
• v.19 no.4
• /
• pp.123-132
• /
• 2013

As the smartphones are equipped with various sensors such as the accelerometer, GPS, gravity sensor, gyros, ambient light sensor, proximity sensor, and so on, there have been many research works on making use of these sensors to create valuable applications. Human activity recognition is one such application that is motivated by various welfare applications such as the support for the elderly, measurement of calorie consumption, analysis of lifestyles, analysis of exercise patterns, and so on. One of the challenges faced when using the smartphone sensors for activity recognition is that the number of sensors used should be minimized to save the battery power. When the number of sensors used are restricted, it is difficult to realize a highly accurate activity recognizer or a classifier because it is hard to distinguish between subtly different activities relying on only limited information. The difficulty gets especially severe when the number of different activity classes to be distinguished is very large. In this paper, we show that a fairly accurate classifier can be built that can distinguish ten different activities by using only a single sensor data, i.e., the smartphone accelerometer data. The approach that we take to dealing with this ten-class problem is to use the ensemble of nested dichotomy (END) method that transforms a multi-class problem into multiple two-class problems. END builds a committee of binary classifiers in a nested fashion using a binary tree. At the root of the binary tree, the set of all the classes are split into two subsets of classes by using a binary classifier. At a child node of the tree, a subset of classes is again split into two smaller subsets by using another binary classifier. Continuing in this way, we can obtain a binary tree where each leaf node contains a single class. This binary tree can be viewed as a nested dichotomy that can make multi-class predictions. Depending on how a set of classes are split into two subsets at each node, the final tree that we obtain can be different. Since there can be some classes that are correlated, a particular tree may perform better than the others. However, we can hardly identify the best tree without deep domain knowledge. The END method copes with this problem by building multiple dichotomy trees randomly during learning, and then combining the predictions made by each tree during classification. The END method is generally known to perform well even when the base learner is unable to model complex decision boundaries As the base classifier at each node of the dichotomy, we have used another ensemble classifier called the random forest. A random forest is built by repeatedly generating a decision tree each time with a different random subset of features using a bootstrap sample. By combining bagging with random feature subset selection, a random forest enjoys the advantage of having more diverse ensemble members than a simple bagging. As an overall result, our ensemble of nested dichotomy can actually be seen as a committee of committees of decision trees that can deal with a multi-class problem with high accuracy. The ten classes of activities that we distinguish in this paper are 'Sitting', 'Standing', 'Walking', 'Running', 'Walking Uphill', 'Walking Downhill', 'Running Uphill', 'Running Downhill', 'Falling', and 'Hobbling'. The features used for classifying these activities include not only the magnitude of acceleration vector at each time point but also the maximum, the minimum, and the standard deviation of vector magnitude within a time window of the last 2 seconds, etc. For experiments to compare the performance of END with those of other methods, the accelerometer data has been collected at every 0.1 second for 2 minutes for each activity from 5 volunteers. Among these 5,900 ($=5{\times}(60{\times}2-2)/0.1$) data collected for each activity (the data for the first 2 seconds are trashed because they do not have time window data), 4,700 have been used for training and the rest for testing. Although 'Walking Uphill' is often confused with some other similar activities, END has been found to classify all of the ten activities with a fairly high accuracy of 98.4%. On the other hand, the accuracies achieved by a decision tree, a k-nearest neighbor, and a one-versus-rest support vector machine have been observed as 97.6%, 96.5%, and 97.6%, respectively.

• Journal of Intelligence and Information Systems
• /
• v.26 no.4
• /
• pp.111-126
• /
• 2020

Most industries have recently become aware of the importance of customer lifetime value as they are exposed to a competitive environment. As a result, preventing customers from churn is becoming a more important business issue than securing new customers. This is because maintaining churn customers is far more economical than securing new customers, and in fact, the acquisition cost of new customers is known to be five to six times higher than the maintenance cost of churn customers. Also, Companies that effectively prevent customer churn and improve customer retention rates are known to have a positive effect on not only increasing the company's profitability but also improving its brand image by improving customer satisfaction. Predicting customer churn, which had been conducted as a sub-research area for CRM, has recently become more important as a big data-based performance marketing theme due to the development of business machine learning technology. Until now, research on customer churn prediction has been carried out actively in such sectors as the mobile telecommunication industry, the financial industry, the distribution industry, and the game industry, which are highly competitive and urgent to manage churn. In addition, These churn prediction studies were focused on improving the performance of the churn prediction model itself, such as simply comparing the performance of various models, exploring features that are effective in forecasting departures, or developing new ensemble techniques, and were limited in terms of practical utilization because most studies considered the entire customer group as a group and developed a predictive model. As such, the main purpose of the existing related research was to improve the performance of the predictive model itself, and there was a relatively lack of research to improve the overall customer churn prediction process. In fact, customers in the business have different behavior characteristics due to heterogeneous transaction patterns, and the resulting churn rate is different, so it is unreasonable to assume the entire customer as a single customer group. Therefore, it is desirable to segment customers according to customer classification criteria, such as loyalty, and to operate an appropriate churn prediction model individually, in order to carry out effective customer churn predictions in heterogeneous industries. Of course, in some studies, there are studies in which customers are subdivided using clustering techniques and applied a churn prediction model for individual customer groups. Although this process of predicting churn can produce better predictions than a single predict model for the entire customer population, there is still room for improvement in that clustering is a mechanical, exploratory grouping technique that calculates distances based on inputs and does not reflect the strategic intent of an entity such as loyalties. This study proposes a segment-based customer departure prediction process (CCP/2DL: Customer Churn Prediction based on Two-Dimensional Loyalty segmentation) based on two-dimensional customer loyalty, assuming that successful customer churn management can be better done through improvements in the overall process than through the performance of the model itself. CCP/2DL is a series of churn prediction processes that segment two-way, quantitative and qualitative loyalty-based customer, conduct secondary grouping of customer segments according to churn patterns, and then independently apply heterogeneous churn prediction models for each churn pattern group. Performance comparisons were performed with the most commonly applied the General churn prediction process and the Clustering-based churn prediction process to assess the relative excellence of the proposed churn prediction process. The General churn prediction process used in this study refers to the process of predicting a single group of customers simply intended to be predicted as a machine learning model, using the most commonly used churn predicting method. And the Clustering-based churn prediction process is a method of first using clustering techniques to segment customers and implement a churn prediction model for each individual group. In cooperation with a global NGO, the proposed CCP/2DL performance showed better performance than other methodologies for predicting churn. This churn prediction process is not only effective in predicting churn, but can also be a strategic basis for obtaining a variety of customer observations and carrying out other related performance marketing activities.

• Journal of the Korean Society for Library and Information Science
• /
• v.15
• /
• pp.225-266
• /
• 1988

Various theories and techniques for the subject catalog have been developed since Charles Ammi Cutter first tried to formulate rules for the construction of subject headings in 1876. However, they do not seem to be appropriate to Korean language because the syntax and semantics of Korean language are different from those of English and other European languages. This study therefore attempts to develop a new Korean subject indexing system, namely Korean String Indexing System(KOSIS), in order to increase the use of subject catalogs. For this purpose, advantages and disadvantages between the classed subject catalog nd the alphabetical subject catalog, which are typical subject ca-alogs in libraries, are investigated, and most of remarkable subject indexing systems, in particular the PRECIS developed by the British National Bibliography, are reviewed and analysed. KOSIS is a string indexing based on purely the syntax and semantics of Korean language, even though considerable principles of PRECIS are applied to it. The outlines of KOSIS are as follows: 1) KOSIS is based on the fundamentals of natural language and an ingenious conjunction of human indexing skills and computer capabilities. 2) KOSIS is. 3 string indexing based on the 'principle of context-dependency.' A string of terms organized accoding to his principle shows remarkable affinity with certain patterns of words in ordinary discourse. From that point onward, natural language rather than classificatory terms become the basic model for indexing schemes. 3) KOSIS uses 24 role operators. One or more operators should be allocated to the index string, which is organized manually by the indexer's intellectual work, in order to establish the most explicit syntactic relationship of index terms. 4) Traditionally, a single -line entry format is used in which a subject heading or index entry is presented as a single sequence of words, consisting of the entry terms, plus, in some cases, an extra qualifying term or phrase. But KOSIS employs a two-line entry format which contains three basic positions for the production of index entries. The 'lead' serves as the user's access point, the 'display' contains those terms which are themselves context dependent on the lead, 'qualifier' sets the lead term into its wider context. 5) Each of the KOSIS entries is co-extensive with the initial subject statement prepared by the indexer, since it displays all the subject specificities. Compound terms are always presented in their natural language order. Inverted headings are not produced in KOSIS. Consequently, the precision ratio of information retrieval can be increased. 6) KOSIS uses 5 relational codes for the system of references among semantically related terms. Semantically related terms are handled by a different set of routines, leading to the production of 'See' and 'See also' references. 7) KOSIS was riginally developed for a classified catalog system which requires a subject index, that is an index -which 'trans-lates' subject index, that is, an index which 'translates' subjects expressed in natural language into the appropriate classification numbers. However, KOSIS can also be us d for a dictionary catalog system. Accordingly, KOSIS strings can be manipulated to produce either appropriate subject indexes for a classified catalog system, or acceptable subject headings for a dictionary catalog system. 8) KOSIS is able to maintain a constistency of index entries and cross references by means of a routine identification of the established index strings and reference system. For this purpose, an individual Subject Indicator Number and Reference Indicator Number is allocated to each new index strings and new index terms, respectively. can produce all the index entries, cross references, and authority cards by means of either manual or mechanical methods. Thus, detailed algorithms for the machine-production of various outputs are provided for the institutions which can use computer facilities.

• Science of Emotion and Sensibility
• /
• v.16 no.1
• /
• pp.107-116
• /
• 2013

In the implementation of a P300 speller, rows and columns paradigm (RCP) is most commonly used. However, the RCP remains subject to adjacency-distraction error and double-flash problems. This study suggests a novel P300 speller stimuli presentation-the sub-block paradigm (SBP) that is likely to solve the problems effectively. Fifteen subjects participated in this experiment where both SBP and RCP were used to implement the P300 speller. Electroencephalography (EEG) activity was recorded from Fz, Cz, Pz, Oz, P3, P4, PO7, and PO8. Each paradigm consisted of a training phase to train a classifier and a testing phase to evaluate the speller. Eighteen characters were used for the target stimuli in the training phase. Additionally, 5 subjects were required to spell 50 characters and the rest of the subjects were to spell 25 characters in the testing phase. Classification accuracy results show that average accuracy was significantly higher in SBP as of 83.73% than that of RCP as of 66.40%. Grand mean event-related potentials (ERPs) at Pz show that positive peak amplitude for the target stimuli was greater in SBP compared to that of RCP. It was found that subjects tended to attend more to the characters in SBP. According to the participants' ratings on how comfortable they were with using each type of paradigm on 7-point Likert scale, most subjects responded 'very difficult' in RCP while responding 'medium' and 'easy' in SBP. The result showed that SBP was felt more comfortable than RCP by the subjects. In sum, the SBP was more correct in P300 speller performance as well as more convenient for users than the RCP. The actual limitations in the study were discussed in the last part of this paper.

• Journal of Intelligence and Information Systems
• /
• v.16 no.4
• /
• pp.213-225
• /
• 2010

Data mining has empowered the managers who are charge of the tasks in their company to present personalized and differentiated marketing programs to their customers with the rapid growth of information technology. Most studies on customer' response have focused on predicting whether they would respond or not for their marketing promotion as marketing managers have been eager to identify who would respond to their marketing promotion. So many studies utilizing data mining have tried to resolve the binary decision problems such as bankruptcy prediction, network intrusion detection, and fraud detection in credit card usages. The prediction of customer's response has been studied with similar methods mentioned above because the prediction of customer's response is a kind of dichotomous decision problem. In addition, a number of competitive data mining techniques such as neural networks, SVM(support vector machine), decision trees, logit, and genetic algorithms have been applied to the prediction of customer's response for marketing promotion. The marketing managers also have tried to classify their customers with quantitative measures such as recency, frequency, and monetary acquired from their transaction database. The measures mean that their customers came to purchase in recent or old days, how frequent in a period, and how much they spent once. Using segmented customers we proposed an approach that could enable to differentiate customers in the same rating among the segmented customers. Our approach employed support vector regression to forecast the purchase amount of customers for each customer rating. Our study used the sample that included 41,924 customers extracted from DMEF04 Data Set, who purchased at least once in the last two years. We classified customers from first rating to fifth rating based on the purchase amount after giving a marketing promotion. Here, we divided customers into first rating who has a large amount of purchase and fifth rating who are non-respondents for the promotion. Our proposed model forecasted the purchase amount of the customers in the same rating and the marketing managers could make a differentiated and personalized marketing program for each customer even though they were belong to the same rating. In addition, we proposed more efficient learning method by separating the learning samples. We employed two learning methods to compare the performance of proposed learning method with general learning method for SVRs. LMW (Learning Method using Whole data for purchasing customers) is a general learning method for forecasting the purchase amount of customers. And we proposed a method, LMS (Learning Method using Separated data for classification purchasing customers), that makes four different SVR models for each class of customers. To evaluate the performance of models, we calculated MAE (Mean Absolute Error) and MAPE (Mean Absolute Percent Error) for each model to predict the purchase amount of customers. In LMW, the overall performance was 0.670 MAPE and the best performance showed 0.327 MAPE. Generally, the performances of the proposed LMS model were analyzed as more superior compared to the performance of the LMW model. In LMS, we found that the best performance was 0.275 MAPE. The performance of LMS was higher than LMW in each class of customers. After comparing the performance of our proposed method LMS to LMW, our proposed model had more significant performance for forecasting the purchase amount of customers in each class. In addition, our approach will be useful for marketing managers when they need to customers for their promotion. Even if customers were belonging to same class, marketing managers could offer customers a differentiated and personalized marketing promotion.

• Journal of Intelligence and Information Systems
• /
• v.16 no.4
• /
• pp.99-112
• /
• 2010

Ensemble learning is a method for improving the performance of classification and prediction algorithms. It is a method for finding a highly accurateclassifier on the training set by constructing and combining an ensemble of weak classifiers, each of which needs only to be moderately accurate on the training set. Ensemble learning has received considerable attention from machine learning and artificial intelligence fields because of its remarkable performance improvement and flexible integration with the traditional learning algorithms such as decision tree (DT), neural networks (NN), and SVM, etc. In those researches, all of DT ensemble studies have demonstrated impressive improvements in the generalization behavior of DT, while NN and SVM ensemble studies have not shown remarkable performance as shown in DT ensembles. Recently, several works have reported that the performance of ensemble can be degraded where multiple classifiers of an ensemble are highly correlated with, and thereby result in multicollinearity problem, which leads to performance degradation of the ensemble. They have also proposed the differentiated learning strategies to cope with performance degradation problem. Hansen and Salamon (1990) insisted that it is necessary and sufficient for the performance enhancement of an ensemble that the ensemble should contain diverse classifiers. Breiman (1996) explored that ensemble learning can increase the performance of unstable learning algorithms, but does not show remarkable performance improvement on stable learning algorithms. Unstable learning algorithms such as decision tree learners are sensitive to the change of the training data, and thus small changes in the training data can yield large changes in the generated classifiers. Therefore, ensemble with unstable learning algorithms can guarantee some diversity among the classifiers. To the contrary, stable learning algorithms such as NN and SVM generate similar classifiers in spite of small changes of the training data, and thus the correlation among the resulting classifiers is very high. This high correlation results in multicollinearity problem, which leads to performance degradation of the ensemble. Kim,s work (2009) showedthe performance comparison in bankruptcy prediction on Korea firms using tradition prediction algorithms such as NN, DT, and SVM. It reports that stable learning algorithms such as NN and SVM have higher predictability than the unstable DT. Meanwhile, with respect to their ensemble learning, DT ensemble shows the more improved performance than NN and SVM ensemble. Further analysis with variance inflation factor (VIF) analysis empirically proves that performance degradation of ensemble is due to multicollinearity problem. It also proposes that optimization of ensemble is needed to cope with such a problem. This paper proposes a hybrid system for coverage optimization of NN ensemble (CO-NN) in order to improve the performance of NN ensemble. Coverage optimization is a technique of choosing a sub-ensemble from an original ensemble to guarantee the diversity of classifiers in coverage optimization process. CO-NN uses GA which has been widely used for various optimization problems to deal with the coverage optimization problem. The GA chromosomes for the coverage optimization are encoded into binary strings, each bit of which indicates individual classifier. The fitness function is defined as maximization of error reduction and a constraint of variance inflation factor (VIF), which is one of the generally used methods to measure multicollinearity, is added to insure the diversity of classifiers by removing high correlation among the classifiers. We use Microsoft Excel and the GAs software package called Evolver. Experiments on company failure prediction have shown that CO-NN is effectively applied in the stable performance enhancement of NNensembles through the choice of classifiers by considering the correlations of the ensemble. The classifiers which have the potential multicollinearity problem are removed by the coverage optimization process of CO-NN and thereby CO-NN has shown higher performance than a single NN classifier and NN ensemble at 1% significance level, and DT ensemble at 5% significance level. However, there remain further research issues. First, decision optimization process to find optimal combination function should be considered in further research. Secondly, various learning strategies to deal with data noise should be introduced in more advanced further researches in the future.

• Journal of Intelligence and Information Systems
• /
• v.19 no.2
• /
• pp.39-54
• /
• 2013

Recent explosive increase of electronic commerce provides many advantageous purchase opportunities to customers. In this situation, customers who do not have enough knowledge about their purchases, may accept product recommendations. Product recommender systems automatically reflect user's preference and provide recommendation list to the users. Thus, product recommender system in online shopping store has been known as one of the most popular tools for one-to-one marketing. However, recommender systems which do not properly reflect user's preference cause user's disappointment and waste of time. In this study, we propose a novel recommender system which uses data mining and multi-model ensemble techniques to enhance the recommendation performance through reflecting the precise user's preference. The research data is collected from the real-world online shopping store, which deals products from famous art galleries and museums in Korea. The data initially contain 5759 transaction data, but finally remain 3167 transaction data after deletion of null data. In this study, we transform the categorical variables into dummy variables and exclude outlier data. The proposed model consists of two steps. The first step predicts customers who have high likelihood to purchase products in the online shopping store. In this step, we first use logistic regression, decision trees, and artificial neural networks to predict customers who have high likelihood to purchase products in each product group. We perform above data mining techniques using SAS E-Miner software. In this study, we partition datasets into two sets as modeling and validation sets for the logistic regression and decision trees. We also partition datasets into three sets as training, test, and validation sets for the artificial neural network model. The validation dataset is equal for the all experiments. Then we composite the results of each predictor using the multi-model ensemble techniques such as bagging and bumping. Bagging is the abbreviation of "Bootstrap Aggregation" and it composite outputs from several machine learning techniques for raising the performance and stability of prediction or classification. This technique is special form of the averaging method. Bumping is the abbreviation of "Bootstrap Umbrella of Model Parameter," and it only considers the model which has the lowest error value. The results show that bumping outperforms bagging and the other predictors except for "Poster" product group. For the "Poster" product group, artificial neural network model performs better than the other models. In the second step, we use the market basket analysis to extract association rules for co-purchased products. We can extract thirty one association rules according to values of Lift, Support, and Confidence measure. We set the minimum transaction frequency to support associations as 5%, maximum number of items in an association as 4, and minimum confidence for rule generation as 10%. This study also excludes the extracted association rules below 1 of lift value. We finally get fifteen association rules by excluding duplicate rules. Among the fifteen association rules, eleven rules contain association between products in "Office Supplies" product group, one rules include the association between "Office Supplies" and "Fashion" product groups, and other three rules contain association between "Office Supplies" and "Home Decoration" product groups. Finally, the proposed product recommender systems provides list of recommendations to the proper customers. We test the usability of the proposed system by using prototype and real-world transaction and profile data. For this end, we construct the prototype system by using the ASP, Java Script and Microsoft Access. In addition, we survey about user satisfaction for the recommended product list from the proposed system and the randomly selected product lists. The participants for the survey are 173 persons who use MSN Messenger, Daum Caf$\acute{e}$, and P2P services. We evaluate the user satisfaction using five-scale Likert measure. This study also performs "Paired Sample T-test" for the results of the survey. The results show that the proposed model outperforms the random selection model with 1% statistical significance level. It means that the users satisfied the recommended product list significantly. The results also show that the proposed system may be useful in real-world online shopping store.

• Journal of Intelligence and Information Systems
• /
• v.25 no.3
• /
• pp.239-251
• /
• 2019

Statistical methods such as moving averages, Kalman filtering, exponential smoothing, regression analysis, and ARIMA (autoregressive integrated moving average) have been used for stock market predictions. However, these statistical methods have not produced superior performances. In recent years, machine learning techniques have been widely used in stock market predictions, including artificial neural network, SVM, and genetic algorithm. In particular, a case-based reasoning method, known as k-nearest neighbor is also widely used for stock price prediction. Case based reasoning retrieves several similar cases from previous cases when a new problem occurs, and combines the class labels of similar cases to create a classification for the new problem. However, case based reasoning has some problems. First, case based reasoning has a tendency to search for a fixed number of neighbors in the observation space and always selects the same number of neighbors rather than the best similar neighbors for the target case. So, case based reasoning may have to take into account more cases even when there are fewer cases applicable depending on the subject. Second, case based reasoning may select neighbors that are far away from the target case. Thus, case based reasoning does not guarantee an optimal pseudo-neighborhood for various target cases, and the predictability can be degraded due to a deviation from the desired similar neighbor. This paper examines how the size of learning data affects stock price predictability through k-nearest neighbor and compares the predictability of k-nearest neighbor with the random walk model according to the size of the learning data and the number of neighbors. In this study, Samsung electronics stock prices were predicted by dividing the learning dataset into two types. For the prediction of next day's closing price, we used four variables: opening value, daily high, daily low, and daily close. In the first experiment, data from January 1, 2000 to December 31, 2017 were used for the learning process. In the second experiment, data from January 1, 2015 to December 31, 2017 were used for the learning process. The test data is from January 1, 2018 to August 31, 2018 for both experiments. We compared the performance of k-NN with the random walk model using the two learning dataset. The mean absolute percentage error (MAPE) was 1.3497 for the random walk model and 1.3570 for the k-NN for the first experiment when the learning data was small. However, the mean absolute percentage error (MAPE) for the random walk model was 1.3497 and the k-NN was 1.2928 for the second experiment when the learning data was large. These results show that the prediction power when more learning data are used is higher than when less learning data are used. Also, this paper shows that k-NN generally produces a better predictive power than random walk model for larger learning datasets and does not when the learning dataset is relatively small. Future studies need to consider macroeconomic variables related to stock price forecasting including opening price, low price, high price, and closing price. Also, to produce better results, it is recommended that the k-nearest neighbor needs to find nearest neighbors using the second step filtering method considering fundamental economic variables as well as a sufficient amount of learning data.

• Korean Journal of Agricultural and Forest Meteorology
• /
• v.23 no.1
• /
• pp.15-33
• /
• 2021

In this study, the possibility of discriminating Fire blight (FB) infection tested using the hyperspectral imagery. The reflectance of healthy and infected leaves and branches was acquired with 5 nm of full width at high maximum (FWHM) and then it was standardized to 10 nm, 25 nm, 50 nm, and 80 nm of FWHM. The standardized samples were divided into training and test sets at ratios of 7:3, 5:5 and 3:7 to find the optimal bands of FWHM by the decision tree analysis. Classification accuracy was evaluated using overall accuracy (OA) and kappa coefficient (KC). The hyperspectral reflectance of infected leaves and branches was significantly lower than those of healthy green, red-edge (RE) and near infrared (NIR) regions. The bands selected for the first node were generally 750 and 800 nm; these were used to identify the infection of leaves and branches, respectively. The accuracy of the classifier was higher in the 7:3 ratio. Four bands with 50 nm of FWHM (450, 650, 750, and 950 nm) might be reasonable because the difference in the recalculated accuracy between 8 bands with 10 nm of FWHM (440, 580, 640, 660, 680, 710, 730, and 740 nm) and 4 bands was only 1.8% for OA and 4.1% for KC, respectively. Finally, adding two bands (550 nm and 800 nm with 25 nm of FWHM) in four bands with 50 nm of FWHM have been proposed to improve the usability of multispectral image sensors with performing various roles in agriculture as well as detecting FB with other combinations of spectral bands.

• Journal of Intelligence and Information Systems
• /
• v.28 no.2
• /
• pp.191-206
• /
• 2022

Recently, it is a de-facto approach to utilize a pre-trained language model(PLM) to achieve the state-of-the-art performance for various natural language tasks(called downstream tasks) such as sentiment analysis and question answering. However, similar to any other machine learning method, PLM tends to depend on the data distribution seen during the training phase and shows worse performance on the unseen (Out-of-Distribution) domain. Due to the aforementioned reason, there have been many efforts to develop domain-specified PLM for various fields such as medical and legal industries. In this paper, we discuss the training of a finance domain-specified PLM for the Korean language and its applications. Our finance domain-specified PLM, KB-BERT, is trained on a carefully curated financial corpus that includes domain-specific documents such as financial reports. We provide extensive performance evaluation results on three natural language tasks, topic classification, sentiment analysis, and question answering. Compared to the state-of-the-art Korean PLM models such as KoELECTRA and KLUE-RoBERTa, KB-BERT shows comparable performance on general datasets based on common corpora like Wikipedia and news articles. Moreover, KB-BERT outperforms compared models on finance domain datasets that require finance-specific knowledge to solve given problems.