• Journal of Intelligence and Information Systems
• /
• v.20 no.4
• /
• pp.121-139
• /
• 2014

Predicting corporate failure has been an important topic in accounting and finance. The costs associated with bankruptcy are high, so the accuracy of bankruptcy prediction is greatly important for financial institutions. Lots of researchers have dealt with the topic associated with bankruptcy prediction in the past three decades. The current research attempts to use ensemble models for improving the performance of bankruptcy prediction. Ensemble classification is to combine individually trained classifiers in order to gain more accurate prediction than individual models. Ensemble techniques are shown to be very useful for improving the generalization ability of the classifier. Bagging is the most commonly used methods for constructing ensemble classifiers. In bagging, the different training data subsets are randomly drawn with replacement from the original training dataset. Base classifiers are trained on the different bootstrap samples. Instance selection is to select critical instances while deleting and removing irrelevant and harmful instances from the original set. Instance selection and bagging are quite well known in data mining. However, few studies have dealt with the integration of instance selection and bagging. This study proposes an improved bagging ensemble based on instance selection using genetic algorithms (GA) for improving the performance of SVM. GA is an efficient optimization procedure based on the theory of natural selection and evolution. GA uses the idea of survival of the fittest by progressively accepting better solutions to the problems. GA searches by maintaining a population of solutions from which better solutions are created rather than making incremental changes to a single solution to the problem. The initial solution population is generated randomly and evolves into the next generation by genetic operators such as selection, crossover and mutation. The solutions coded by strings are evaluated by the fitness function. The proposed model consists of two phases: GA based Instance Selection and Instance based Bagging. In the first phase, GA is used to select optimal instance subset that is used as input data of bagging model. In this study, the chromosome is encoded as a form of binary string for the instance subset. In this phase, the population size was set to 100 while maximum number of generations was set to 150. We set the crossover rate and mutation rate to 0.7 and 0.1 respectively. We used the prediction accuracy of model as the fitness function of GA. SVM model is trained on training data set using the selected instance subset. The prediction accuracy of SVM model over test data set is used as fitness value in order to avoid overfitting. In the second phase, we used the optimal instance subset selected in the first phase as input data of bagging model. We used SVM model as base classifier for bagging ensemble. The majority voting scheme was used as a combining method in this study. This study applies the proposed model to the bankruptcy prediction problem using a real data set from Korean companies. The research data used in this study contains 1832 externally non-audited firms which filed for bankruptcy (916 cases) and non-bankruptcy (916 cases). Financial ratios categorized as stability, profitability, growth, activity and cash flow were investigated through literature review and basic statistical methods and we selected 8 financial ratios as the final input variables. We separated the whole data into three subsets as training, test and validation data set. In this study, we compared the proposed model with several comparative models including the simple individual SVM model, the simple bagging model and the instance selection based SVM model. The McNemar tests were used to examine whether the proposed model significantly outperforms the other models. The experimental results show that the proposed model outperforms the other models.

• Journal of National Security and Military Science
• /
• s.9
• /
• pp.313-340
• /
• 2011

The National Defense Reformation bill of "National Defense Reformation 2020" which have been constantly disputed and reformed by the government went through various levels of complementary measures after the North Korean sinking on the Republic of Korea (ROK) Naval Vessel "Cheonan". The final outcome of this reform is also known as the 307 Plan and this was announced on the 8th March. The reformed National Defense Reformation is to reduce the number of units and military personnel under the military structure reformation. However, in order for us to undertake successful National Defense Reformation, the use of privatized civilian resources are essential. Therefore according to this theory, the ROK Ministry of National Defense (MND) have selected the usage of privatized resources as one of the main core agenda for the National Defense Reformation management procedures, and under this agenda the MND plans to further expand the usage of private Especially the MND plans to minimize the personnel resources applied in non-combat areas and in turn use these supplemented personnel with optimization. In order to do this, the MND have initiated necessary appropriate analysis over the whole national defense section by understanding various projects and acquisition requests required by each militaries and civilian research institutions. However for efficient management of privatized civilian resources, first of all, those possible efficient private resources which can achieve optimization will need to be identified, and secondly continuous systematic reinforcements will need to be made in private resource usage legislations. Furthermore, we would need to consider the possibility of labor disputes because of privatization expansion. Therefore, full legal and systematic complementary measures are required in all possible issue arising areas which can affect the combat readiness posture. There is another problem of huge increase in operational expenses as reduction of standby forces are only reducing the number of soldiers and filling these numbers with more cost expensive commissioned officers. However, to overcome this problem, we would need to reduce the number of positions available for active officers and fill these positions with military reserve personnel who previously had working experiences with the related positions (thereby guaranteeing active officers re-employment after completing active service). This would in tum maintain the standards of combat readiness posture and reduce necessary financial budgets which may newly arise. The area of maintenance, supply, transportation, training & education duties which are highly efficient when using privatized resources, will need to be transformed from military management based to civilian management based system. For maintenance, this can be processed by integrating National Maintenance Support System. In order for us to undertake this procedure, we would need to develop maintenance units which are possible to be privatized and this will in turn reduce the military personnel executing job duties, improve service quality and prevent duplicate investments etc. For supply area, we will need to establish Integrated Military Logistics Center in-connection with national and civilian logistics system. This will in turn reduce the logistics time frame as well as required personnel and equipments. In terms of transportation, we will need to further expand the renting and leasing system. This will need to be executed by integrating the National Defense Transportation Information System which will in turn reduce the required personnel and financial budgets. Finally for training and education, retired military personnel can be employed as training instructors and at the military academy, further expansion in the number of civilian professors can be employed in-connection with National Defense Reformation. In other words, more active privatized civilian resources will need to be managed and used for National Defense Reformation.

• Journal of Intelligence and Information Systems
• /
• v.22 no.2
• /
• pp.127-142
• /
• 2016

Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.

• Journal of Korea Water Resources Association
• /
• v.54 no.10
• /
• pp.795-805
• /
• 2021

In this study, after developing an LSTM-based deep learning model for estimating daily runoff in the Soyang River Dam basin, the accuracy of the model for various combinations of model structure and input data was investigated. A model was built based on the database consisting of average daily precipitation, average daily temperature, average daily wind speed (input up to here), and daily average flow rate (output) during the first 12 years (1997.1.1-2008.12.31). The Nash-Sutcliffe Model Efficiency Coefficient (NSE) and RMSE were examined for validation using the flow discharge data of the later 12 years (2009.1.1-2020.12.31). The combination that showed the highest accuracy was the case in which all possible input data (12 years of daily precipitation, weather temperature, wind speed) were used on the LSTM model structure with 64 hidden units. The NSE and RMSE of the verification period were 0.862 and 76.8 m³/s, respectively. When the number of hidden units of LSTM exceeds 500, the performance degradation of the model due to overfitting begins to appear, and when the number of hidden units exceeds 1000, the overfitting problem becomes prominent. A model with very high performance (NSE=0.8~0.84) could be obtained when only 12 years of daily precipitation was used for model training. A model with reasonably high performance (NSE=0.63-0.85) when only one year of input data was used for model training. In particular, an accurate model (NSE=0.85) could be obtained if the one year of training data contains a wide magnitude of flow events such as extreme flow and droughts as well as normal events. If the training data includes both the normal and extreme flow rates, input data that is longer than 5 years did not significantly improve the model performance.

• Asia Marketing Journal
• /
• v.16 no.1
• /
• pp.71-100
• /
• 2014

While the role of emotional antecedents of effective selling behavior would be important, the issue has not been fully addressed in the sales literature. To fill this gap, we conceptualize and empirically examine the relationships among salesperson's emotional regulation processes such as emotional intelligence (EI) and emotional labor (EL), effective selling behavior, and sales performance on the basis of educational, occupational, social psychology literature and marketing literature (e.g., Henning-Thurau, Groth, Paul, and Gremler 2006; Kidwell et al. 2011; Liu et al. 2008; Mayer, Salovey, and Caruso 2008). First, salesperson's EI is defined as his or her capability that enables correct perceptions about emotional situations in sales interactions. The EI is expected to work as psychological resources for different types of EL (i.e., deep acting and surface acting) to be performed by salesperson as emotional expression strategies (e.g., Lie et al. 2008). It is, then, expected that the features of EL selected by the salesperson would lead to different levels of adaptive selling behavior (ASB) and thereby sales performance (Monaghan 2006). Further, given that salesperson's customer orientation (CO) is found to be an important correlate of ASB (Franke and Park 2006), it is expected that CO would moderate the relationship between EL and ASB (Rozell, Pettijohn, and Parker 2004). Hence, this research attempts to shed additional light on emotionally-driven (EL) as well as cognitively-driven (CO) antecedents of ASB (Frank and Park 2006). The findings of the survey research, done with 336 salespersons in insurance and financial companies, are summarized as follows. First, salespersons with a high level of EI are found to use both deep acting (regulating the emotions themselves) and surface acting (controlling only emotional expressions) in a versatile way, when implementing EL. Second, the more the salesperson performs deep acting, the more he or she shows ASB. It is, then, important for salespersons to use deep acting more frequently in the EL process in order to enhance the quality of interacting with customers through ASB. On the other hand, the salesperson's surface acting did not have a significant relationship with ASB. Moreover, CO was found to moderate the relationship between the salesperson's deep acting and ASB. That is, the context of high CO culture and individual salesperson's deep acting would synergistically make the selling efforts adaptive to customer preferences. Conceptualizing and empirically verifying the antecedent roles of important emotional constructs such as EI and EL in salesperson's effective selling behavior (ASB) and sales performance is a major theoretical contribution in the sales literature. Managerially, this research provides a deeper understanding on the nature of tasks performed by salespersons in service industries and a few guidelines for managing the sales force. First, sales organizations had better consciously assess EI capacity in the selection and nurturing processes of salespersons, given that EI can efficiently drive EL and the resulting effective selling behavior and performance. Further, the concept of EL could provide a framework to understand the salespersons' emotional experiences in depth. Especially, sales organizations may well think over how to develop deep acting capabilities of their sales representatives. In this direction, the training on deep acting strategies would be an essential task for improving effective selling behavior and performance of salespersons. This kind of training had better incorporate the perspectives of customers such that many customers can actually discern whether salespersons are doing either surface acting or deep acting. Finally, based on the synergistic effects of deep acting and CO culture, how to build and sustain CO is always an ever-important task in sales organizations. While the prior sales literature has emphasized the process and structure of highly customer-oriented sales organization, our research not only corroborates the important aspects of customer-oriented sales organization, but also adds the important dimension of competent sales representatives who can resonate with customers by deep acting for sales excellence.

• Journal of Intelligence and Information Systems
• /
• v.26 no.1
• /
• pp.1-21
• /
• 2020

With the rapid development of artificial intelligence technology, various techniques have been developed to extract meaningful information from unstructured text data which constitutes a large portion of big data. Over the past decades, text mining technologies have been utilized in various industries for practical applications. In the field of business intelligence, it has been employed to discover new market and/or technology opportunities and support rational decision making of business participants. The market information such as market size, market growth rate, and market share is essential for setting companies' business strategies. There has been a continuous demand in various fields for specific product level-market information. However, the information has been generally provided at industry level or broad categories based on classification standards, making it difficult to obtain specific and proper information. In this regard, we propose a new methodology that can estimate the market sizes of product groups at more detailed levels than that of previously offered. We applied Word2Vec algorithm, a neural network based semantic word embedding model, to enable automatic market size estimation from individual companies' product information in a bottom-up manner. The overall process is as follows: First, the data related to product information is collected, refined, and restructured into suitable form for applying Word2Vec model. Next, the preprocessed data is embedded into vector space by Word2Vec and then the product groups are derived by extracting similar products names based on cosine similarity calculation. Finally, the sales data on the extracted products is summated to estimate the market size of the product groups. As an experimental data, text data of product names from Statistics Korea's microdata (345,103 cases) were mapped in multidimensional vector space by Word2Vec training. We performed parameters optimization for training and then applied vector dimension of 300 and window size of 15 as optimized parameters for further experiments. We employed index words of Korean Standard Industry Classification (KSIC) as a product name dataset to more efficiently cluster product groups. The product names which are similar to KSIC indexes were extracted based on cosine similarity. The market size of extracted products as one product category was calculated from individual companies' sales data. The market sizes of 11,654 specific product lines were automatically estimated by the proposed model. For the performance verification, the results were compared with actual market size of some items. The Pearson's correlation coefficient was 0.513. Our approach has several advantages differing from the previous studies. First, text mining and machine learning techniques were applied for the first time on market size estimation, overcoming the limitations of traditional sampling based- or multiple assumption required-methods. In addition, the level of market category can be easily and efficiently adjusted according to the purpose of information use by changing cosine similarity threshold. Furthermore, it has a high potential of practical applications since it can resolve unmet needs for detailed market size information in public and private sectors. Specifically, it can be utilized in technology evaluation and technology commercialization support program conducted by governmental institutions, as well as business strategies consulting and market analysis report publishing by private firms. The limitation of our study is that the presented model needs to be improved in terms of accuracy and reliability. The semantic-based word embedding module can be advanced by giving a proper order in the preprocessed dataset or by combining another algorithm such as Jaccard similarity with Word2Vec. Also, the methods of product group clustering can be changed to other types of unsupervised machine learning algorithm. Our group is currently working on subsequent studies and we expect that it can further improve the performance of the conceptually proposed basic model in this study.

• Journal of Intelligence and Information Systems
• /
• v.18 no.1
• /
• pp.125-141
• /
• 2012

These days, the malicious attacks and hacks on the networked systems are dramatically increasing, and the patterns of them are changing rapidly. Consequently, it becomes more important to appropriately handle these malicious attacks and hacks, and there exist sufficient interests and demand in effective network security systems just like intrusion detection systems. Intrusion detection systems are the network security systems for detecting, identifying and responding to unauthorized or abnormal activities appropriately. Conventional intrusion detection systems have generally been designed using the experts' implicit knowledge on the network intrusions or the hackers' abnormal behaviors. However, they cannot handle new or unknown patterns of the network attacks, although they perform very well under the normal situation. As a result, recent studies on intrusion detection systems use artificial intelligence techniques, which can proactively respond to the unknown threats. For a long time, researchers have adopted and tested various kinds of artificial intelligence techniques such as artificial neural networks, decision trees, and support vector machines to detect intrusions on the network. However, most of them have just applied these techniques singularly, even though combining the techniques may lead to better detection. With this reason, we propose a new integrated model for intrusion detection. Our model is designed to combine prediction results of four different binary classification models-logistic regression (LOGIT), decision trees (DT), artificial neural networks (ANN), and support vector machines (SVM), which may be complementary to each other. As a tool for finding optimal combining weights, genetic algorithms (GA) are used. Our proposed model is designed to be built in two steps. At the first step, the optimal integration model whose prediction error (i.e. erroneous classification rate) is the least is generated. After that, in the second step, it explores the optimal classification threshold for determining intrusions, which minimizes the total misclassification cost. To calculate the total misclassification cost of intrusion detection system, we need to understand its asymmetric error cost scheme. Generally, there are two common forms of errors in intrusion detection. The first error type is the False-Positive Error (FPE). In the case of FPE, the wrong judgment on it may result in the unnecessary fixation. The second error type is the False-Negative Error (FNE) that mainly misjudges the malware of the program as normal. Compared to FPE, FNE is more fatal. Thus, total misclassification cost is more affected by FNE rather than FPE. To validate the practical applicability of our model, we applied it to the real-world dataset for network intrusion detection. The experimental dataset was collected from the IDS sensor of an official institution in Korea from January to June 2010. We collected 15,000 log data in total, and selected 10,000 samples from them by using random sampling method. Also, we compared the results from our model with the results from single techniques to confirm the superiority of the proposed model. LOGIT and DT was experimented using PASW Statistics v18.0, and ANN was experimented using Neuroshell R4.0. For SVM, LIBSVM v2.90-a freeware for training SVM classifier-was used. Empirical results showed that our proposed model based on GA outperformed all the other comparative models in detecting network intrusions from the accuracy perspective. They also showed that the proposed model outperformed all the other comparative models in the total misclassification cost perspective. Consequently, it is expected that our study may contribute to build cost-effective intelligent intrusion detection systems.

• Journal of Intelligence and Information Systems
• /
• v.23 no.1
• /
• pp.95-108
• /
• 2017

Recently, AlphaGo which is Bakuk (Go) artificial intelligence program by Google DeepMind, had a huge victory against Lee Sedol. Many people thought that machines would not be able to win a man in Go games because the number of paths to make a one move is more than the number of atoms in the universe unlike chess, but the result was the opposite to what people predicted. After the match, artificial intelligence technology was focused as a core technology of the fourth industrial revolution and attracted attentions from various application domains. Especially, deep learning technique have been attracted as a core artificial intelligence technology used in the AlphaGo algorithm. The deep learning technique is already being applied to many problems. Especially, it shows good performance in image recognition field. In addition, it shows good performance in high dimensional data area such as voice, image and natural language, which was difficult to get good performance using existing machine learning techniques. However, in contrast, it is difficult to find deep leaning researches on traditional business data and structured data analysis. In this study, we tried to find out whether the deep learning techniques have been studied so far can be used not only for the recognition of high dimensional data but also for the binary classification problem of traditional business data analysis such as customer churn analysis, marketing response prediction, and default prediction. And we compare the performance of the deep learning techniques with that of traditional artificial neural network models. The experimental data in the paper is the telemarketing response data of a bank in Portugal. It has input variables such as age, occupation, loan status, and the number of previous telemarketing and has a binary target variable that records whether the customer intends to open an account or not. In this study, to evaluate the possibility of utilization of deep learning algorithms and techniques in binary classification problem, we compared the performance of various models using CNN, LSTM algorithm and dropout, which are widely used algorithms and techniques in deep learning, with that of MLP models which is a traditional artificial neural network model. However, since all the network design alternatives can not be tested due to the nature of the artificial neural network, the experiment was conducted based on restricted settings on the number of hidden layers, the number of neurons in the hidden layer, the number of output data (filters), and the application conditions of the dropout technique. The F1 Score was used to evaluate the performance of models to show how well the models work to classify the interesting class instead of the overall accuracy. The detail methods for applying each deep learning technique in the experiment is as follows. The CNN algorithm is a method that reads adjacent values from a specific value and recognizes the features, but it does not matter how close the distance of each business data field is because each field is usually independent. In this experiment, we set the filter size of the CNN algorithm as the number of fields to learn the whole characteristics of the data at once, and added a hidden layer to make decision based on the additional features. For the model having two LSTM layers, the input direction of the second layer is put in reversed position with first layer in order to reduce the influence from the position of each field. In the case of the dropout technique, we set the neurons to disappear with a probability of 0.5 for each hidden layer. The experimental results show that the predicted model with the highest F1 score was the CNN model using the dropout technique, and the next best model was the MLP model with two hidden layers using the dropout technique. In this study, we were able to get some findings as the experiment had proceeded. First, models using dropout techniques have a slightly more conservative prediction than those without dropout techniques, and it generally shows better performance in classification. Second, CNN models show better classification performance than MLP models. This is interesting because it has shown good performance in binary classification problems which it rarely have been applied to, as well as in the fields where it's effectiveness has been proven. Third, the LSTM algorithm seems to be unsuitable for binary classification problems because the training time is too long compared to the performance improvement. From these results, we can confirm that some of the deep learning algorithms can be applied to solve business binary classification problems.

• Journal of Intelligence and Information Systems
• /
• v.27 no.3
• /
• pp.139-156
• /
• 2021

The biggest reason for using a deep learning model in image classification is that it is possible to consider the relationship between each region by extracting each region's features from the overall information of the image. However, the CNN model may not be suitable for emotional image data without the image's regional features. To solve the difficulty of classifying emotion images, many researchers each year propose a CNN-based architecture suitable for emotion images. Studies on the relationship between color and human emotion were also conducted, and results were derived that different emotions are induced according to color. In studies using deep learning, there have been studies that apply color information to image subtraction classification. The case where the image's color information is additionally used than the case where the classification model is trained with only the image improves the accuracy of classifying image emotions. This study proposes two ways to increase the accuracy by incorporating the result value after the model classifies an image's emotion. Both methods improve accuracy by modifying the result value based on statistics using the color of the picture. When performing the test by finding the two-color combinations most distributed for all training data, the two-color combinations most distributed for each test data image were found. The result values were corrected according to the color combination distribution. This method weights the result value obtained after the model classifies an image's emotion by creating an expression based on the log function and the exponential function. Emotion6, classified into six emotions, and Artphoto classified into eight categories were used for the image data. Densenet169, Mnasnet, Resnet101, Resnet152, and Vgg19 architectures were used for the CNN model, and the performance evaluation was compared before and after applying the two-stage learning to the CNN model. Inspired by color psychology, which deals with the relationship between colors and emotions, when creating a model that classifies an image's sentiment, we studied how to improve accuracy by modifying the result values based on color. Sixteen colors were used: red, orange, yellow, green, blue, indigo, purple, turquoise, pink, magenta, brown, gray, silver, gold, white, and black. It has meaning. Using Scikit-learn's Clustering, the seven colors that are primarily distributed in the image are checked. Then, the RGB coordinate values of the colors from the image are compared with the RGB coordinate values of the 16 colors presented in the above data. That is, it was converted to the closest color. Suppose three or more color combinations are selected. In that case, too many color combinations occur, resulting in a problem in which the distribution is scattered, so a situation fewer influences the result value. Therefore, to solve this problem, two-color combinations were found and weighted to the model. Before training, the most distributed color combinations were found for all training data images. The distribution of color combinations for each class was stored in a Python dictionary format to be used during testing. During the test, the two-color combinations that are most distributed for each test data image are found. After that, we checked how the color combinations were distributed in the training data and corrected the result. We devised several equations to weight the result value from the model based on the extracted color as described above. The data set was randomly divided by 80:20, and the model was verified using 20% of the data as a test set. After splitting the remaining 80% of the data into five divisions to perform 5-fold cross-validation, the model was trained five times using different verification datasets. Finally, the performance was checked using the test dataset that was previously separated. Adam was used as the activation function, and the learning rate was set to 0.01. The training was performed as much as 20 epochs, and if the validation loss value did not decrease during five epochs of learning, the experiment was stopped. Early tapping was set to load the model with the best validation loss value. The classification accuracy was better when the extracted information using color properties was used together than the case using only the CNN architecture.

• Asia pacific journal of information systems
• /
• v.18 no.4
• /
• pp.105-130
• /
• 2008

Consumers of e-learning market differ from those of other markets in that they are replaced in a specific time scale. For example, e-learning contents aimed at highschool senior students cannot be consumed by a specific consumer over the designated period of time. Hence e-learning service providers need to attract new groups of students every year. Due to lack of information on products designed for continuously emerging consumers, the consumers face difficulties in making rational decisions in a short time period. Increased uncertainty of product purchase leads customers to herding behaviors to obtain information of the product from others and imitate them. Taking into consideration of these features of e-learning market, this study will focus on the online herding behavior in purchasing e-learning contents. There is no definite concept for e-learning. However, it is being discussed in a wide range of perspectives from educational engineering to management to e-business etc. Based upon the existing studies, we identify two main view-points regarding e-learning. The first defines e-learning as a concept that includes existing terminologies, such as CBT (Computer Based Training), WBT (Web Based Training), and IBT (Internet Based Training). In this view, e-learning utilizes IT in order to support professors and a part of or entire education systems. In the second perspective, e-learning is defined as the usage of Internet technology to deliver diverse intelligence and achievement enhancing solutions. In other words, only the educations that are done through the Internet and network can be classified as e-learning. We take the second definition of e-learning for our working definition. The main goal of this study is to investigate what factors affect consumer intention to purchase e-learning contents and to identify the differential impact of the factors between consumers with purchase experience and those without the experience. To accomplish the goal of this study, it focuses on herding behavior and perceived usefulness as antecedents to behavioral intention. The proposed research model in the study extends the Technology Acceptance Model by adding herding behavior and usability to take into account the unique characteristics of e-learning content market and e-learning systems use, respectively. The current study also includes consumer experience with e-learning content purchase because the previous experience is believed to affect purchasing intention when consumers buy experience goods or services. Previous studies on e-learning did not consider the characteristics of e-learning contents market and the differential impact of consumer experience on the relationship between the antecedents and behavioral intention, which is the target of this study. This study employs a survey method to empirically test the proposed research model. A survey questionnaire was developed and distributed to 629 informants. 528 responses were collected, which consist of potential customer group (n = 133) and experienced customer group (n = 395). The data were analyzed using PLS method, a structural equation modeling method. Overall, both herding behavior and perceived usefulness influence consumer intention to purchase e-learning contents. In detail, in the case of potential customer group, herding behavior has stronger effect on purchase intention than does perceived usefulness. However, in the case of shopping-experienced customer group, perceived usefulness has stronger effect than does herding behavior. In sum, the results of the analysis show that with regard to purchasing experience, perceived usefulness and herding behavior had differential effects upon the purchase of e-learning contents. As a follow-up analysis, the interaction effects of the number of purchase transaction and herding behavior/perceived usefulness on purchase intention were investigated. The results show that there are no interaction effects. This study contributes to the literature in a couple of ways. From a theoretical perspective, this study examined and showed evidence that the characteristics of e-learning market such as continuous renewal of consumers and thus high uncertainty and individual experiences are important factors to be considered when the purchase intention of e-learning content is studied. This study can be used as a basis for future studies on e-learning success. From a practical perspective, this study provides several important implications on what types of marketing strategies e-learning companies need to build. The bottom lines of these strategies include target group attraction, word-of-mouth management, enhancement of web site usability quality, etc. The limitations of this study are also discussed for future studies.