• Journal of Intelligence and Information Systems
• /
• v.23 no.2
• /
• pp.71-88
• /
• 2017

Language models were originally developed for speech recognition and language processing. Using a set of example sentences, a language model predicts the next word or character based on sequential input data. N-gram models have been widely used but this model cannot model the correlation between the input units efficiently since it is a probabilistic model which are based on the frequency of each unit in the training set. Recently, as the deep learning algorithm has been developed, a recurrent neural network (RNN) model and a long short-term memory (LSTM) model have been widely used for the neural language model (Ahn, 2016; Kim et al., 2016; Lee et al., 2016). These models can reflect dependency between the objects that are entered sequentially into the model (Gers and Schmidhuber, 2001; Mikolov et al., 2010; Sundermeyer et al., 2012). In order to learning the neural language model, texts need to be decomposed into words or morphemes. Since, however, a training set of sentences includes a huge number of words or morphemes in general, the size of dictionary is very large and so it increases model complexity. In addition, word-level or morpheme-level models are able to generate vocabularies only which are contained in the training set. Furthermore, with highly morphological languages such as Turkish, Hungarian, Russian, Finnish or Korean, morpheme analyzers have more chance to cause errors in decomposition process (Lankinen et al., 2016). Therefore, this paper proposes a phoneme-level language model for Korean language based on LSTM models. A phoneme such as a vowel or a consonant is the smallest unit that comprises Korean texts. We construct the language model using three or four LSTM layers. Each model was trained using Stochastic Gradient Algorithm and more advanced optimization algorithms such as Adagrad, RMSprop, Adadelta, Adam, Adamax, and Nadam. Simulation study was done with Old Testament texts using a deep learning package Keras based the Theano. After pre-processing the texts, the dataset included 74 of unique characters including vowels, consonants, and punctuation marks. Then we constructed an input vector with 20 consecutive characters and an output with a following 21st character. Finally, total 1,023,411 sets of input-output vectors were included in the dataset and we divided them into training, validation, testsets with proportion 70:15:15. All the simulation were conducted on a system equipped with an Intel Xeon CPU (16 cores) and a NVIDIA GeForce GTX 1080 GPU. We compared the loss function evaluated for the validation set, the perplexity evaluated for the test set, and the time to be taken for training each model. As a result, all the optimization algorithms but the stochastic gradient algorithm showed similar validation loss and perplexity, which are clearly superior to those of the stochastic gradient algorithm. The stochastic gradient algorithm took the longest time to be trained for both 3- and 4-LSTM models. On average, the 4-LSTM layer model took 69% longer training time than the 3-LSTM layer model. However, the validation loss and perplexity were not improved significantly or became even worse for specific conditions. On the other hand, when comparing the automatically generated sentences, the 4-LSTM layer model tended to generate the sentences which are closer to the natural language than the 3-LSTM model. Although there were slight differences in the completeness of the generated sentences between the models, the sentence generation performance was quite satisfactory in any simulation conditions: they generated only legitimate Korean letters and the use of postposition and the conjugation of verbs were almost perfect in the sense of grammar. The results of this study are expected to be widely used for the processing of Korean language in the field of language processing and speech recognition, which are the basis of artificial intelligence systems.

• Journal of Intelligence and Information Systems
• /
• v.18 no.3
• /
• pp.79-96
• /
• 2012

As the pace of competition dramatically accelerates and the complexity of change grows, a variety of research have been conducted to improve firms' short-term performance and to enhance firms' long-term survival. In particular, researchers and practitioners have paid their attention to identify promising technologies that lead competitive advantage to a firm. Discovery of promising technology depends on how a firm evaluates the value of technologies, thus many evaluating methods have been proposed. Experts' opinion based approaches have been widely accepted to predict the value of technologies. Whereas this approach provides in-depth analysis and ensures validity of analysis results, it is usually cost-and time-ineffective and is limited to qualitative evaluation. Considerable studies attempt to forecast the value of technology by using patent information to overcome the limitation of experts' opinion based approach. Patent based technology evaluation has served as a valuable assessment approach of the technological forecasting because it contains a full and practical description of technology with uniform structure. Furthermore, it provides information that is not divulged in any other sources. Although patent information based approach has contributed to our understanding of prediction of promising technologies, it has some limitations because prediction has been made based on the past patent information, and the interpretations of patent analyses are not consistent. In order to fill this gap, this study proposes a technology forecasting methodology by integrating patent information approach and artificial intelligence method. The methodology consists of three modules : evaluation of technologies promising, implementation of technologies value prediction model, and recommendation of promising technologies. In the first module, technologies promising is evaluated from three different and complementary dimensions; impact, fusion, and diffusion perspectives. The impact of technologies refers to their influence on future technologies development and improvement, and is also clearly associated with their monetary value. The fusion of technologies denotes the extent to which a technology fuses different technologies, and represents the breadth of search underlying the technology. The fusion of technologies can be calculated based on technology or patent, thus this study measures two types of fusion index; fusion index per technology and fusion index per patent. Finally, the diffusion of technologies denotes their degree of applicability across scientific and technological fields. In the same vein, diffusion index per technology and diffusion index per patent are considered respectively. In the second module, technologies value prediction model is implemented using artificial intelligence method. This studies use the values of five indexes (i.e., impact index, fusion index per technology, fusion index per patent, diffusion index per technology and diffusion index per patent) at different time (e.g., t-n, t-n-1, t-n-2, ${\cdots}$) as input variables. The out variables are values of five indexes at time t, which is used for learning. The learning method adopted in this study is backpropagation algorithm. In the third module, this study recommends final promising technologies based on analytic hierarchy process. AHP provides relative importance of each index, leading to final promising index for technology. Applicability of the proposed methodology is tested by using U.S. patents in international patent class G06F (i.e., electronic digital data processing) from 2000 to 2008. The results show that mean absolute error value for prediction produced by the proposed methodology is lower than the value produced by multiple regression analysis in cases of fusion indexes. However, mean absolute error value of the proposed methodology is slightly higher than the value of multiple regression analysis. These unexpected results may be explained, in part, by small number of patents. Since this study only uses patent data in class G06F, number of sample patent data is relatively small, leading to incomplete learning to satisfy complex artificial intelligence structure. In addition, fusion index per technology and impact index are found to be important criteria to predict promising technology. This study attempts to extend the existing knowledge by proposing a new methodology for prediction technology value by integrating patent information analysis and artificial intelligence network. It helps managers who want to technology develop planning and policy maker who want to implement technology policy by providing quantitative prediction methodology. In addition, this study could help other researchers by proving a deeper understanding of the complex technological forecasting field.

• Asia pacific journal of information systems
• /
• v.17 no.4
• /
• pp.31-59
• /
• 2007

We frequently use search engines to find relevant information in the Web but still end up with too much information. In order to solve this problem of information overload, ranking algorithms have been applied to various domains. As more information will be available in the future, effectively and efficiently ranking search results will become more critical. In this paper, we propose a ranking algorithm for the Semantic Web resources, specifically RDF resources. Traditionally, the importance of a particular Web page is estimated based on the number of key words found in the page, which is subject to manipulation. In contrast, link analysis methods such as Google's PageRank capitalize on the information which is inherent in the link structure of the Web graph. PageRank considers a certain page highly important if it is referred to by many other pages. The degree of the importance also increases if the importance of the referring pages is high. Kleinberg's algorithm is another link-structure based ranking algorithm for Web pages. Unlike PageRank, Kleinberg's algorithm utilizes two kinds of scores: the authority score and the hub score. If a page has a high authority score, it is an authority on a given topic and many pages refer to it. A page with a high hub score links to many authoritative pages. As mentioned above, the link-structure based ranking method has been playing an essential role in World Wide Web(WWW), and nowadays, many people recognize the effectiveness and efficiency of it. On the other hand, as Resource Description Framework(RDF) data model forms the foundation of the Semantic Web, any information in the Semantic Web can be expressed with RDF graph, making the ranking algorithm for RDF knowledge bases greatly important. The RDF graph consists of nodes and directional links similar to the Web graph. As a result, the link-structure based ranking method seems to be highly applicable to ranking the Semantic Web resources. However, the information space of the Semantic Web is more complex than that of WWW. For instance, WWW can be considered as one huge class, i.e., a collection of Web pages, which has only a recursive property, i.e., a 'refers to' property corresponding to the hyperlinks. However, the Semantic Web encompasses various kinds of classes and properties, and consequently, ranking methods used in WWW should be modified to reflect the complexity of the information space in the Semantic Web. Previous research addressed the ranking problem of query results retrieved from RDF knowledge bases. Mukherjea and Bamba modified Kleinberg's algorithm in order to apply their algorithm to rank the Semantic Web resources. They defined the objectivity score and the subjectivity score of a resource, which correspond to the authority score and the hub score of Kleinberg's, respectively. They concentrated on the diversity of properties and introduced property weights to control the influence of a resource on another resource depending on the characteristic of the property linking the two resources. A node with a high objectivity score becomes the object of many RDF triples, and a node with a high subjectivity score becomes the subject of many RDF triples. They developed several kinds of Semantic Web systems in order to validate their technique and showed some experimental results verifying the applicability of their method to the Semantic Web. Despite their efforts, however, there remained some limitations which they reported in their paper. First, their algorithm is useful only when a Semantic Web system represents most of the knowledge pertaining to a certain domain. In other words, the ratio of links to nodes should be high, or overall resources should be described in detail, to a certain degree for their algorithm to properly work. Second, a Tightly-Knit Community(TKC) effect, the phenomenon that pages which are less important but yet densely connected have higher scores than the ones that are more important but sparsely connected, remains as problematic. Third, a resource may have a high score, not because it is actually important, but simply because it is very common and as a consequence it has many links pointing to it. In this paper, we examine such ranking problems from a novel perspective and propose a new algorithm which can solve the problems under the previous studies. Our proposed method is based on a class-oriented approach. In contrast to the predicate-oriented approach entertained by the previous research, a user, under our approach, determines the weights of a property by comparing its relative significance to the other properties when evaluating the importance of resources in a specific class. This approach stems from the idea that most queries are supposed to find resources belonging to the same class in the Semantic Web, which consists of many heterogeneous classes in RDF Schema. This approach closely reflects the way that people, in the real world, evaluate something, and will turn out to be superior to the predicate-oriented approach for the Semantic Web. Our proposed algorithm can resolve the TKC(Tightly Knit Community) effect, and further can shed lights on other limitations posed by the previous research. In addition, we propose two ways to incorporate data-type properties which have not been employed even in the case when they have some significance on the resource importance. We designed an experiment to show the effectiveness of our proposed algorithm and the validity of ranking results, which was not tried ever in previous research. We also conducted a comprehensive mathematical analysis, which was overlooked in previous research. The mathematical analysis enabled us to simplify the calculation procedure. Finally, we summarize our experimental results and discuss further research issues.

• Journal of Intelligence and Information Systems
• /
• v.24 no.1
• /
• pp.101-124
• /
• 2018

Recently, most of the technologies have been developed in various forms through the advancement of single technology or interaction with other technologies. Particularly, these technologies have the characteristic of the convergence caused by the interaction between two or more techniques. In addition, efforts in responding to technological changes by advance are continuously increasing through forecasting promising convergence technologies that will emerge in the near future. According to this phenomenon, many researchers are attempting to perform various analyses about forecasting promising convergence technologies. A convergence technology has characteristics of various technologies according to the principle of generation. Therefore, forecasting promising convergence technologies is much more difficult than forecasting general technologies with high growth potential. Nevertheless, some achievements have been confirmed in an attempt to forecasting promising technologies using big data analysis and social network analysis. Studies of convergence technology through data analysis are actively conducted with the theme of discovering new convergence technologies and analyzing their trends. According that, information about new convergence technologies is being provided more abundantly than in the past. However, existing methods in analyzing convergence technology have some limitations. Firstly, most studies deal with convergence technology analyze data through predefined technology classifications. The technologies appearing recently tend to have characteristics of convergence and thus consist of technologies from various fields. In other words, the new convergence technologies may not belong to the defined classification. Therefore, the existing method does not properly reflect the dynamic change of the convergence phenomenon. Secondly, in order to forecast the promising convergence technologies, most of the existing analysis method use the general purpose indicators in process. This method does not fully utilize the specificity of convergence phenomenon. The new convergence technology is highly dependent on the existing technology, which is the origin of that technology. Based on that, it can grow into the independent field or disappear rapidly, according to the change of the dependent technology. In the existing analysis, the potential growth of convergence technology is judged through the traditional indicators designed from the general purpose. However, these indicators do not reflect the principle of convergence. In other words, these indicators do not reflect the characteristics of convergence technology, which brings the meaning of new technologies emerge through two or more mature technologies and grown technologies affect the creation of another technology. Thirdly, previous studies do not provide objective methods for evaluating the accuracy of models in forecasting promising convergence technologies. In the studies of convergence technology, the subject of forecasting promising technologies was relatively insufficient due to the complexity of the field. Therefore, it is difficult to find a method to evaluate the accuracy of the model that forecasting promising convergence technologies. In order to activate the field of forecasting promising convergence technology, it is important to establish a method for objectively verifying and evaluating the accuracy of the model proposed by each study. To overcome these limitations, we propose a new method for analysis of convergence technologies. First of all, through topic modeling, we derive a new technology classification in terms of text content. It reflects the dynamic change of the actual technology market, not the existing fixed classification standard. In addition, we identify the influence relationships between technologies through the topic correspondence weights of each document, and structuralize them into a network. In addition, we devise a centrality indicator (PGC, potential growth centrality) to forecast the future growth of technology by utilizing the centrality information of each technology. It reflects the convergence characteristics of each technology, according to technology maturity and interdependence between technologies. Along with this, we propose a method to evaluate the accuracy of forecasting model by measuring the growth rate of promising technology. It is based on the variation of potential growth centrality by period. In this paper, we conduct experiments with 13,477 patent documents dealing with technical contents to evaluate the performance and practical applicability of the proposed method. As a result, it is confirmed that the forecast model based on a centrality indicator of the proposed method has a maximum forecast accuracy of about 2.88 times higher than the accuracy of the forecast model based on the currently used network indicators.

• KOREAN JOURNAL OF CROP SCIENCE
• /
• v.34 no.s02
• /
• pp.45-65
• /
• 1989

The Korean peninsular having the complexity of the photography and variability of climate is located within passing area of a lots of typhoon occurring from the southern islands of Philippines. So, there are various patterns of wind and flooding damages in paddy field occuring by the strong wind and the heavy rain concentrated during the summer season of rice growing period in Korea. The wind damages to rice plants in Korea were mainly caused by saline wind, dry wind and strong wind when typhoon occurred. The saline wind damage having symptom of white head or dried leaves occurred by 1.1 to 17.2 mg of salt per dry weight stuck on the plant which was located at 2. 5km away from seashore of southern coastal area during the period(from 27th to 29th, August, 1986) of typhoon &Vera& accompanying 62-96％ of relative humidity, more than 6 m per second of wind velocity and 22.5 to 26.4$^{\circ}C$ of air temperature without rain. Most of the typhoons accompanying 4.0 to 8. 5m per second of wind and low humidity (lesp an 60％) with high temperature in the east coastal area and southen area of Korea. were changed to dry and hot wind by the foehn phenomenon. The dry wind damages with the symptom of the white head or the discolored brownish grain occurred at the rice heading stage. The strong wind caused the severe damages such as the broken leaves, cut-leaves and dried leaves before heading stage, lodging and shattering of grain at ripening stage mechanically during typhoon. To reduce the wind damages to rice plant, cultivation of resistant varieties to wind damages such as Sangpoongbyeo and Cheongcheongbyeo and the escape of heading stage during period of typhoon by accelerating of heading within 15th, August are effective. Though the flood disasters to rice plant such as earring away of field, burying of field, submerging and lodging damage are getting low by the construction of dam for multiple purpose and river bank, they are occasionally occurred by the regional heavy rain and water filled out in bank around the river. Paddy field were submerged for 2 to 4 days when typhoon and heavy rain occurred about the end of August. At this time, the rice plants that was in younger growing stage in the late transplanting field of southern area of Korea had the severe damages. Although panicles of rice plant which was in the meiotic growing stage and heading stage were died when flooded, they had 66％ of yield compensating ability by the upper tilling panicle produced from tiller with dead panicle in ordinary transplanting paddy field. It is effective for reduction of flooding damages to cultivate the resistant variety to flooding having the resistance to bacterial leaf blight, lodging and small brown planthopper simultaneously. Especially, Tongil type rice varieties are relatively resistant to flooding, compared to Japonica rice varieties. Tongil type rice varieties had high survivals, low elongation ability of leaf sheath and blade, high recovering ability by the high root activity and photosynthesis and high yield compensating ability by the upper tillering panicle when flooded. To minimize the flooding and wind damage to rice plants in future, following research have to be carried out; 1. Data analysis by telemetering and computerization of climate, actual conditions and growing diagnosis of crops damaged by disasters. 2. Development of tolerant varieties to poor natural conditions related to flooding and wind damages. 3. Improvement of the reasonable cropping system by introduction of other crops compensating the loss of the damaged rice. 4. Increament of utilization of rice plant which was damaged.

• Journal of Intelligence and Information Systems
• /
• v.25 no.1
• /
• pp.85-107
• /
• 2019

Online consumers browse products belonging to a particular product line or brand for purchase, or simply leave a wide range of navigation without making purchase. The research on the behavior and purchase of online consumers has been steadily progressed, and related services and applications based on behavior data of consumers have been developed in practice. In recent years, customization strategies and recommendation systems of consumers have been utilized due to the development of big data technology, and attempts are being made to optimize users' shopping experience. However, even in such an attempt, it is very unlikely that online consumers will actually be able to visit the website and switch to the purchase stage. This is because online consumers do not just visit the website to purchase products but use and browse the websites differently according to their shopping motives and purposes. Therefore, it is important to analyze various types of visits as well as visits to purchase, which is important for understanding the behaviors of online consumers. In this study, we explored the clustering analysis of session based on click stream data of e-commerce company in order to explain diversity and complexity of search behavior of online consumers and typified search behavior. For the analysis, we converted data points of more than 8 million pages units into visit units' sessions, resulting in a total of over 500,000 website visit sessions. For each visit session, 12 characteristics such as page view, duration, search diversity, and page type concentration were extracted for clustering analysis. Considering the size of the data set, we performed the analysis using the Mini-Batch K-means algorithm, which has advantages in terms of learning speed and efficiency while maintaining the clustering performance similar to that of the clustering algorithm K-means. The most optimized number of clusters was derived from four, and the differences in session unit characteristics and purchasing rates were identified for each cluster. The online consumer visits the website several times and learns about the product and decides the purchase. In order to analyze the purchasing process over several visits of the online consumer, we constructed the visiting sequence data of the consumer based on the navigation patterns in the web site derived clustering analysis. The visit sequence data includes a series of visiting sequences until one purchase is made, and the items constituting one sequence become cluster labels derived from the foregoing. We have separately established a sequence data for consumers who have made purchases and data on visits for consumers who have only explored products without making purchases during the same period of time. And then sequential pattern mining was applied to extract frequent patterns from each sequence data. The minimum support is set to 10%, and frequent patterns consist of a sequence of cluster labels. While there are common derived patterns in both sequence data, there are also frequent patterns derived only from one side of sequence data. We found that the consumers who made purchases through the comparative analysis of the extracted frequent patterns showed the visiting pattern to decide to purchase the product repeatedly while searching for the specific product. The implication of this study is that we analyze the search type of online consumers by using large - scale click stream data and analyze the patterns of them to explain the behavior of purchasing process with data-driven point. Most studies that typology of online consumers have focused on the characteristics of the type and what factors are key in distinguishing that type. In this study, we carried out an analysis to type the behavior of online consumers, and further analyzed what order the types could be organized into one another and become a series of search patterns. In addition, online retailers will be able to try to improve their purchasing conversion through marketing strategies and recommendations for various types of visit and will be able to evaluate the effect of the strategy through changes in consumers' visit patterns.

• Journal of Intelligence and Information Systems
• /
• v.24 no.1
• /
• pp.205-225
• /
• 2018

Convolutional Neural Network (ConvNet) is one class of the powerful Deep Neural Network that can analyze and learn hierarchies of visual features. Originally, first neural network (Neocognitron) was introduced in the 80s. At that time, the neural network was not broadly used in both industry and academic field by cause of large-scale dataset shortage and low computational power. However, after a few decades later in 2012, Krizhevsky made a breakthrough on ILSVRC-12 visual recognition competition using Convolutional Neural Network. That breakthrough revived people interest in the neural network. The success of Convolutional Neural Network is achieved with two main factors. First of them is the emergence of advanced hardware (GPUs) for sufficient parallel computation. Second is the availability of large-scale datasets such as ImageNet (ILSVRC) dataset for training. Unfortunately, many new domains are bottlenecked by these factors. For most domains, it is difficult and requires lots of effort to gather large-scale dataset to train a ConvNet. Moreover, even if we have a large-scale dataset, training ConvNet from scratch is required expensive resource and time-consuming. These two obstacles can be solved by using transfer learning. Transfer learning is a method for transferring the knowledge from a source domain to new domain. There are two major Transfer learning cases. First one is ConvNet as fixed feature extractor, and the second one is Fine-tune the ConvNet on a new dataset. In the first case, using pre-trained ConvNet (such as on ImageNet) to compute feed-forward activations of the image into the ConvNet and extract activation features from specific layers. In the second case, replacing and retraining the ConvNet classifier on the new dataset, then fine-tune the weights of the pre-trained network with the backpropagation. In this paper, we focus on using multiple ConvNet layers as a fixed feature extractor only. However, applying features with high dimensional complexity that is directly extracted from multiple ConvNet layers is still a challenging problem. We observe that features extracted from multiple ConvNet layers address the different characteristics of the image which means better representation could be obtained by finding the optimal combination of multiple ConvNet layers. Based on that observation, we propose to employ multiple ConvNet layer representations for transfer learning instead of a single ConvNet layer representation. Overall, our primary pipeline has three steps. Firstly, images from target task are given as input to ConvNet, then that image will be feed-forwarded into pre-trained AlexNet, and the activation features from three fully connected convolutional layers are extracted. Secondly, activation features of three ConvNet layers are concatenated to obtain multiple ConvNet layers representation because it will gain more information about an image. When three fully connected layer features concatenated, the occurring image representation would have 9192 (4096+4096+1000) dimension features. However, features extracted from multiple ConvNet layers are redundant and noisy since they are extracted from the same ConvNet. Thus, a third step, we will use Principal Component Analysis (PCA) to select salient features before the training phase. When salient features are obtained, the classifier can classify image more accurately, and the performance of transfer learning can be improved. To evaluate proposed method, experiments are conducted in three standard datasets (Caltech-256, VOC07, and SUN397) to compare multiple ConvNet layer representations against single ConvNet layer representation by using PCA for feature selection and dimension reduction. Our experiments demonstrated the importance of feature selection for multiple ConvNet layer representation. Moreover, our proposed approach achieved 75.6% accuracy compared to 73.9% accuracy achieved by FC7 layer on the Caltech-256 dataset, 73.1% accuracy compared to 69.2% accuracy achieved by FC8 layer on the VOC07 dataset, 52.2% accuracy compared to 48.7% accuracy achieved by FC7 layer on the SUN397 dataset. We also showed that our proposed approach achieved superior performance, 2.8%, 2.1% and 3.1% accuracy improvement on Caltech-256, VOC07, and SUN397 dataset respectively compare to existing work.

• Journal of Intelligence and Information Systems
• /
• v.25 no.4
• /
• pp.1-33
• /
• 2019

The small and medium sized enterprises (hereinafter SMEs) are already at a competitive disadvantaged when compared to large companies with more abundant resources. Manufacturing SMEs not only need a lot of information needed for new product development for sustainable growth and survival, but also seek networking to overcome the limitations of resources, but they are faced with limitations due to their size limitations. In a new era in which connectivity increases the complexity and uncertainty of the business environment, SMEs are increasingly urged to find information and solve networking problems. In order to solve these problems, the government funded research institutes plays an important role and duty to solve the information asymmetry problem of SMEs. The purpose of this study is to identify the differentiating characteristics of SMEs that utilize the public information support infrastructure provided by SMEs to enhance the innovation capacity of SMEs, and how they contribute to corporate performance. We argue that we need an infrastructure for providing information support to SMEs as part of this effort to strengthen of the role of government funded institutions; in this study, we specifically identify the target of such a policy and furthermore empirically demonstrate the effects of such policy-based efforts. Our goal is to help establish the strategies for building the information supporting infrastructure. To achieve this purpose, we first classified the characteristics of SMEs that have been found to utilize the information supporting infrastructure provided by government funded institutions. This allows us to verify whether selection bias appears in the analyzed group, which helps us clarify the interpretative limits of our study results. Next, we performed mediator and moderator effect analysis for multiple variables to analyze the process through which the use of information supporting infrastructure led to an improvement in external networking capabilities and resulted in enhancing product competitiveness. This analysis helps identify the key factors we should focus on when offering indirect support to SMEs through the information supporting infrastructure, which in turn helps us more efficiently manage research related to SME supporting policies implemented by government funded institutions. The results of this study showed the following. First, SMEs that used the information supporting infrastructure were found to have a significant difference in size in comparison to domestic R&D SMEs, but on the other hand, there was no significant difference in the cluster analysis that considered various variables. Based on these findings, we confirmed that SMEs that use the information supporting infrastructure are superior in size, and had a relatively higher distribution of companies that transact to a greater degree with large companies, when compared to the SMEs composing the general group of SMEs. Also, we found that companies that already receive support from the information infrastructure have a high concentration of companies that need collaboration with government funded institution. Secondly, among the SMEs that use the information supporting infrastructure, we found that increasing external networking capabilities contributed to enhancing product competitiveness, and while this was no the effect of direct assistance, we also found that indirect contributions were made by increasing the open marketing capabilities: in other words, this was the result of an indirect-only mediator effect. Also, the number of times the company received additional support in this process through mentoring related to information utilization was found to have a mediated moderator effect on improving external networking capabilities and in turn strengthening product competitiveness. The results of this study provide several insights that will help establish policies. KISTI's information support infrastructure may lead to the conclusion that marketing is already well underway, but it intentionally supports groups that enable to achieve good performance. As a result, the government should provide clear priorities whether to support the companies in the underdevelopment or to aid better performance. Through our research, we have identified how public information infrastructure contributes to product competitiveness. Here, we can draw some policy implications. First, the public information support infrastructure should have the capability to enhance the ability to interact with or to find the expert that provides required information. Second, if the utilization of public information support (online) infrastructure is effective, it is not necessary to continuously provide informational mentoring, which is a parallel offline support. Rather, offline support such as mentoring should be used as an appropriate device for abnormal symptom monitoring. Third, it is required that SMEs should improve their ability to utilize, because the effect of enhancing networking capacity through public information support infrastructure and enhancing product competitiveness through such infrastructure appears in most types of companies rather than in specific SMEs.