• Journal of Intelligence and Information Systems
• /
• v.18 no.1
• /
• pp.23-38
• /
• 2012

There is a large difference between purchasing patterns in an online shopping mall and in an offline market. This difference may be caused mainly by the difference in accessibility of online and offline markets. It means that an interval between the initial purchasing decision and its realization appears to be relatively short in an online shopping mall, because a customer can make an order immediately. Because of the short interval between a purchasing decision and its realization, an online shopping mall transaction usually contains fewer items than that of an offline market. In an offline market, customers usually keep some items in mind and buy them all at once a few days after deciding to buy them, instead of buying each item individually and immediately. On the contrary, more than 70% of online shopping mall transactions contain only one item. This statistic implies that traditional data mining techniques cannot be directly applied to online market analysis, because hardly any association rules can survive with an acceptable level of Support because of too many Null Transactions. Most market basket analyses on online shopping mall transactions, therefore, have been performed by expanding the co-occurrence criteria of traditional association rule mining. While the traditional co-occurrence criteria defines items purchased in one transaction as concurrently purchased items, the expanded co-occurrence criteria regards items purchased by a customer during some predefined period (e.g., a day) as concurrently purchased items. In studies using expanded co-occurrence criteria, however, the criteria has been defined arbitrarily by researchers without any theoretical grounds or agreement. The lack of clear grounds of adopting a certain co-occurrence criteria degrades the reliability of the analytical results. Moreover, it is hard to derive new meaningful findings by combining the outcomes of previous individual studies. In this paper, we attempt to compare expanded co-occurrence criteria and propose a guideline for selecting an appropriate one. First of all, we compare the accuracy of association rules discovered according to various co-occurrence criteria. By doing this experiment we expect that we can provide a guideline for selecting appropriate co-occurrence criteria that corresponds to the purpose of the analysis. Additionally, we will perform similar experiments with several groups of customers that are segmented by each customer's average duration between orders. By this experiment, we attempt to discover the relationship between the optimal co-occurrence criteria and the customer's average duration between orders. Finally, by a series of experiments, we expect that we can provide basic guidelines for developing customized recommendation systems. Our experiments use a real dataset acquired from one of the largest internet shopping malls in Korea. We use 66,278 transactions of 3,847 customers conducted during the last two years. Overall results show that the accuracy of association rules of frequent shoppers (whose average duration between orders is relatively short) is higher than that of causal shoppers. In addition we discover that with frequent shoppers, the accuracy of association rules appears very high when the co-occurrence criteria of the training set corresponds to the validation set (i.e., target set). It implies that the co-occurrence criteria of frequent shoppers should be set according to the application purpose period. For example, an analyzer should use a day as a co-occurrence criterion if he/she wants to offer a coupon valid only for a day to potential customers who will use the coupon. On the contrary, an analyzer should use a month as a co-occurrence criterion if he/she wants to publish a coupon book that can be used for a month. In the case of causal shoppers, the accuracy of association rules appears to not be affected by the period of the application purposes. The accuracy of the causal shoppers' association rules becomes higher when the longer co-occurrence criterion has been adopted. It implies that an analyzer has to set the co-occurrence criterion for as long as possible, regardless of the application purpose period.

• Journal of Intelligence and Information Systems
• /
• v.18 no.2
• /
• pp.157-175
• /
• 2012

It is difficult to evaluate the performance of process innovation of e-procurement which has large scale and complex processes. The existing evaluation methods for measuring the effects of process innovation have been mainly done with statistically quantitative methods by analyzing operational data or with qualitative methods by conducting surveys and interviews. However, these methods have some limitations to evaluate the effects because the performance evaluation of e-procurement process innovation should consider the interactions among participants who are active either directly or indirectly through the processes. This study considers the e-procurement process as a complex system and develops a simulation model based on MAS(Multi-Agent System) to evaluate the effects of e-procurement process innovation. Multi-agent based simulation allows observing interaction patterns of objects in virtual world through relationship among objects and their behavioral mechanism. Agent-based simulation is suitable especially for complex business problems. In this study, we used Netlogo Version 4.1.3 as a MAS simulation tool which was developed in Northwestern University. To do this, we developed a interaction model of agents in MAS environment. We defined process agents and task agents, and assigned their behavioral characteristics. The developed simulation model was applied to G2B system (KONEPS: Korea ON-line E-Procurement System) of Public Procurement Service (PPS) in Korea and used to evaluate the innovation effects of the G2B system. KONEPS is a successfully established e-procurement system started in the year 2002. KONEPS is a representative e-Procurement system which integrates characteristics of e-commerce into government for business procurement activities. KONEPS deserves the international recognition considering the annual transaction volume of 56 billion dollars, daily exchanges of electronic documents, users consisted of 121,000 suppliers and 37,000 public organizations, and the 4.5 billion dollars of cost saving. For the simulation, we analyzed the e-procurement of process of KONEPS into eight sub processes such as 'process 1: search products and acquisition of proposal', 'process 2 : review the methods of contracts and item features', 'process 3 : a notice of bid', 'process 4 : registration and confirmation of qualification', 'process 5 : bidding', 'process 6 : a screening test', 'process 7 : contracts', and 'process 8 : invoice and payment'. For the parameter settings of the agents behavior, we collected some data from the transactional database of PPS and some information by conducting a survey. The used data for the simulation are 'participants (government organizations, local government organizations and public institutions)', 'the number of bidding per year', 'the number of total contracts', 'the number of shopping mall transactions', 'the rate of contracts between bidding and shopping mall', 'the successful bidding ratio', and the estimated time for each process. The comparison was done for the difference of time consumption between 'before the innovation (As-was)' and 'after the innovation (As-is).' The results showed that there were productivity improvements in every eight sub processes. The decrease ratio of 'average number of task processing' was 92.7% and the decrease ratio of 'average time of task processing' was 95.4% in entire processes when we use G2B system comparing to the conventional method. Also, this study found that the process innovation effect will be enhanced if the task process related to the 'contract' can be improved. This study shows the usability and possibility of using MAS in process innovation evaluation and its modeling.

• Journal of Intelligence and Information Systems
• /
• v.18 no.2
• /
• pp.61-83
• /
• 2012

A knowledge map describes the network of related knowledge into the form of a diagram, and therefore underpins the structure of knowledge categorizing and archiving by defining the relationship of the referential navigation between knowledge. The referential navigation between knowledge means the relationship of cross-referencing exhibited when a piece of knowledge is utilized by a user. To understand the contents of the knowledge, a user usually requires additionally information or knowledge related with each other in the relation of cause and effect. This relation can be expanded as the effective connection between knowledge increases, and finally forms the network of knowledge. A network display of knowledge using nodes and links to arrange and to represent the relationship between concepts can provide a more complex knowledge structure than a hierarchical display. Moreover, it can facilitate a user to infer through the links shown on the network. For this reason, building a knowledge map based on the ontology technology has been emphasized to formally as well as objectively describe the knowledge and its relationships. As the necessity to build a knowledge map based on the structure of the ontology has been emphasized, not a few researches have been proposed to fulfill the needs. However, most of those researches to apply the ontology to build the knowledge map just focused on formally expressing knowledge and its relationships with other knowledge to promote the possibility of knowledge reuse. Although many types of knowledge maps based on the structure of the ontology were proposed, no researches have tried to design and implement the referential navigation-enabled knowledge map. This paper addresses a methodology to build the ontology-based knowledge map enabling the referential navigation between knowledge. The ontology-based knowledge map resulted from the proposed methodology can not only express the referential navigation between knowledge but also infer additional relationships among knowledge based on the referential relationships. The most highlighted benefits that can be delivered by applying the ontology technology to the knowledge map include; formal expression about knowledge and its relationships with others, automatic identification of the knowledge network based on the function of self-inference on the referential relationships, and automatic expansion of the knowledge-base designed to categorize and store knowledge according to the network between knowledge. To enable the referential navigation between knowledge included in the knowledge map, and therefore to form the knowledge map in the format of a network, the ontology must describe knowledge according to the relation with the process and task. A process is composed of component tasks, while a task is activated after any required knowledge is inputted. Since the relation of cause and effect between knowledge can be inherently determined by the sequence of tasks, the referential relationship between knowledge can be circuitously implemented if the knowledge is modeled to be one of input or output of each task. To describe the knowledge with respect to related process and task, the Protege-OWL, an editor that enables users to build ontologies for the Semantic Web, is used. An OWL ontology-based knowledge map includes descriptions of classes (process, task, and knowledge), properties (relationships between process and task, task and knowledge), and their instances. Given such an ontology, the OWL formal semantics specifies how to derive its logical consequences, i.e. facts not literally present in the ontology, but entailed by the semantics. Therefore a knowledge network can be automatically formulated based on the defined relationships, and the referential navigation between knowledge is enabled. To verify the validity of the proposed concepts, two real business process-oriented knowledge maps are exemplified: the knowledge map of the process of 'Business Trip Application' and 'Purchase Management'. By applying the 'DL-Query' provided by the Protege-OWL as a plug-in module, the performance of the implemented ontology-based knowledge map has been examined. Two kinds of queries to check whether the knowledge is networked with respect to the referential relations as well as the ontology-based knowledge network can infer further facts that are not literally described were tested. The test results show that not only the referential navigation between knowledge has been correctly realized, but also the additional inference has been accurately performed.

• Journal of Intelligence and Information Systems
• /
• v.16 no.3
• /
• pp.121-145
• /
• 2010

University timetabling depending on the educational environments of universities is an NP-hard problem that the amount of computation required to find solutions increases exponentially with the problem size. For many years, there have been lots of studies on university timetabling from the necessity of automatic timetable generation for students' convenience and effective lesson, and for the effective allocation of subjects, lecturers, and classrooms. Timetables are classified into a course timetable and an examination timetable. This study focuses on the former. In general, a course timetable for liberal arts is scheduled by the office of academic affairs and a course timetable for major subjects is scheduled by each department of a university. We found several problems from the analysis of current course timetabling in departments. First, it is time-consuming and inefficient for each department to do the routine and repetitive timetabling work manually. Second, many classes are concentrated into several time slots in a timetable. This tendency decreases the effectiveness of students' classes. Third, several major subjects might overlap some required subjects in liberal arts at the same time slots in the timetable. In this case, it is required that students should choose only one from the overlapped subjects. Fourth, many subjects are lectured by same lecturers every year and most of lecturers prefer the same time slots for the subjects compared with last year. This means that it will be helpful if departments reuse the previous timetables. To solve such problems and support the effective course timetabling in each department, this study proposes a university timetabling support system based on two phases. In the first phase, each department generates a timetable template from the most similar timetable case, which is based on case-based reasoning. In the second phase, the department schedules a timetable with the help of interactive user interface under the timetabling criteria, which is based on rule-based approach. This study provides the illustrations of Hanshin University. We classified timetabling criteria into intrinsic and extrinsic criteria. In intrinsic criteria, there are three criteria related to lecturer, class, and classroom which are all hard constraints. In extrinsic criteria, there are four criteria related to 'the numbers of lesson hours' by the lecturer, 'prohibition of lecture allocation to specific day-hours' for committee members, 'the number of subjects in the same day-hour,' and 'the use of common classrooms.' In 'the numbers of lesson hours' by the lecturer, there are three kinds of criteria : 'minimum number of lesson hours per week,' 'maximum number of lesson hours per week,' 'maximum number of lesson hours per day.' Extrinsic criteria are also all hard constraints except for 'minimum number of lesson hours per week' considered as a soft constraint. In addition, we proposed two indices for measuring similarities between subjects of current semester and subjects of the previous timetables, and for evaluating distribution degrees of a scheduled timetable. Similarity is measured by comparison of two attributes-subject name and its lecturer-between current semester and a previous semester. The index of distribution degree, based on information entropy, indicates a distribution of subjects in the timetable. To show this study's viability, we implemented a prototype system and performed experiments with the real data of Hanshin University. Average similarity from the most similar cases of all departments was estimated as 41.72%. It means that a timetable template generated from the most similar case will be helpful. Through sensitivity analysis, the result shows that distribution degree will increase if we set 'the number of subjects in the same day-hour' to more than 90%.

• Journal of Intelligence and Information Systems
• /
• v.16 no.3
• /
• pp.99-120
• /
• 2010

Due to the wide spread of customers' frequent access of non face-to-face services, there have been many attempts to improve customer satisfaction using huge amounts of data accumulated throughnon face-to-face channels. Usually, a call center is regarded to be one of the most representative non-faced channels. Therefore, it is important that a call center has enough agents to offer high level customer satisfaction. However, managing too many agents would increase the operational costs of a call center by increasing labor costs. Therefore, predicting and calculating the appropriate size of human resources of a call center is one of the most critical success factors of call center management. For this reason, most call centers are currently establishing a department of WFM(Work Force Management) to estimate the appropriate number of agents and to direct much effort to predict the volume of inbound calls. In real world applications, inbound call prediction is usually performed based on the intuition and experience of a domain expert. In other words, a domain expert usually predicts the volume of calls by calculating the average call of some periods and adjusting the average according tohis/her subjective estimation. However, this kind of approach has radical limitations in that the result of prediction might be strongly affected by the expert's personal experience and competence. It is often the case that a domain expert may predict inbound calls quite differently from anotherif the two experts have mutually different opinions on selecting influential variables and priorities among the variables. Moreover, it is almost impossible to logically clarify the process of expert's subjective prediction. Currently, to overcome the limitations of subjective call prediction, most call centers are adopting a WFMS(Workforce Management System) package in which expert's best practices are systemized. With WFMS, a user can predict the volume of calls by calculating the average call of each day of the week, excluding some eventful days. However, WFMS costs too much capital during the early stage of system establishment. Moreover, it is hard to reflect new information ontothe system when some factors affecting the amount of calls have been changed. In this paper, we attempt to devise a new model for predicting inbound calls that is not only based on theoretical background but also easily applicable to real world applications. Our model was mainly developed by the interactive decision tree technique, one of the most popular techniques in data mining. Therefore, we expect that our model can predict inbound calls automatically based on historical data, and it can utilize expert's domain knowledge during the process of tree construction. To analyze the accuracy of our model, we performed intensive experiments on a real case of one of the largest car insurance companies in Korea. In the case study, the prediction accuracy of the devised two models and traditional WFMS are analyzed with respect to the various error rates allowable. The experiments reveal that our data mining-based two models outperform WFMS in terms of predicting the amount of accident calls and fault calls in most experimental situations examined.

• Journal of Intelligence and Information Systems
• /
• v.25 no.4
• /
• pp.67-88
• /
• 2019

Webtoon is a Korean-style digital comics platform that distributes comics content produced using the characteristic elements of the Internet in a form that can be consumed online. With the recent rapid growth of the webtoon industry and the exponential increase in the supply of webtoon content, the need for effective webtoon content recommendation measures is growing. Webtoons are digital content products that combine pictorial, literary and digital elements. Therefore, webtoons stimulate consumer sentiment by making readers have fun and engaging and empathizing with the situations in which webtoons are produced. In this context, it can be expected that the sentiment that webtoons evoke to consumers will serve as an important criterion for consumers' choice of webtoons. However, there is a lack of research to improve webtoons' recommendation performance by utilizing consumer sentiment. This study is aimed at developing consumer sentiment pattern maps that can support effective recommendations of webtoon content, focusing on consumer sentiments that have not been fully discussed previously. Metadata and consumer sentiments data were collected for 200 works serviced on the Korean webtoon platform 'Naver Webtoon' to conduct this study. 488 sentiment terms were collected for 127 works, excluding those that did not meet the purpose of the analysis. Next, similar or duplicate terms were combined or abstracted in accordance with the bottom-up approach. As a result, we have built webtoons specialized sentiment-index, which are reduced to a total of 63 emotive adjectives. By performing exploratory factor analysis on the constructed sentiment-index, we have derived three important dimensions for classifying webtoon types. The exploratory factor analysis was performed through the Principal Component Analysis (PCA) using varimax factor rotation. The three dimensions were named 'Immersion', 'Touch' and 'Irritant' respectively. Based on this, K-Means clustering was performed and the entire webtoons were classified into four types. Each type was named 'Snack', 'Drama', 'Irritant', and 'Romance'. For each type of webtoon, we wrote webtoon-sentiment 2-Mode network graphs and looked at the characteristics of the sentiment pattern appearing for each type. In addition, through profiling analysis, we were able to derive meaningful strategic implications for each type of webtoon. First, The 'Snack' cluster is a collection of webtoons that are fast-paced and highly entertaining. Many consumers are interested in these webtoons, but they don't rate them well. Also, consumers mostly use simple expressions of sentiment when talking about these webtoons. Webtoons belonging to 'Snack' are expected to appeal to modern people who want to consume content easily and quickly during short travel time, such as commuting time. Secondly, webtoons belonging to 'Drama' are expected to evoke realistic and everyday sentiments rather than exaggerated and light comic ones. When consumers talk about webtoons belonging to a 'Drama' cluster in online, they are found to express a variety of sentiments. It is appropriate to establish an OSMU(One source multi-use) strategy to extend these webtoons to other content such as movies and TV series. Third, the sentiment pattern map of 'Irritant' shows the sentiments that discourage customer interest by stimulating discomfort. Webtoons that evoke these sentiments are hard to get public attention. Artists should pay attention to these sentiments that cause inconvenience to consumers in creating webtoons. Finally, Webtoons belonging to 'Romance' do not evoke a variety of consumer sentiments, but they are interpreted as touching consumers. They are expected to be consumed as 'healing content' targeted at consumers with high levels of stress or mental fatigue in their lives. The results of this study are meaningful in that it identifies the applicability of consumer sentiment in the areas of recommendation and classification of webtoons, and provides guidelines to help members of webtoons' ecosystem better understand consumers and formulate strategies.

• 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 Intelligence and Information Systems
• /
• v.27 no.3
• /
• pp.113-137
• /
• 2021

According to the clear potential of mobile banking growth, many studies related to this are being conducted, but in Korea, it is concentrated on the analysis of technical factors or consumers' intentions, behaviors, and satisfaction. In addition, even though it has a strong customer base of 20s, there are few studies that have been conducted specifically for this customer group. In order for mobile banking to take a leap forward, a strategy to secure various perspectives is needed not only through research on itself but also through research on external factors affecting mobile banking. Therefore, this study analyzes impulsiveness, credit card use, and SNS addiction among various external factors that can significantly affect mobile banking in their 20s. This study examines whether the relationship between impulsiveness and mobile banking usage depends on whether or not a credit card is used, and checks whether a customer's impulsiveness is possible by examining whether a credit card is used. Based on this, it is possible to establish new standards for classification of marketing target groups of mobile banking. After finding out the static or unsuitable relationship between whether to use a credit card and impulsiveness, we want to indirectly predict the customer's impulsiveness through whether to use a credit card or not to use a credit card. It also verifies the mediating effect of SNS addiction in the relationship between impulsiveness and mobile banking usage. For this analysis, the collected data were conducted according to research problems using the SPSS Statistics 25 program. The findings are as follows. First, positive urgency has been shown to have a significant static effect on mobile banking usage. Second, whether to use credit cards has shown moderating effects in the relationship between fraudulent urgency and mobile banking usage. Third, it has been shown that all subfactors of impulsiveness have significant static relationships with subfactors of SNS addiction. Fourth, it has been confirmed that the relationship between positive urgency, SNS addiction, and mobile banking usage has total effect and direct effect. The first result means that mobile banking usage may be high if positive urgency is measured relatively high, even if the multi-dimensional impulsiveness scale is low. The second result indicates that mobile banking usage rates were not affected by the independent variable, negative urgency, but were found to have a significant static relationship with negative urgency when using credit cards. The third result means that SNS is likely to become addictive if lack of premeditation or lack of perseverance is high because it provides instant enjoyment and satisfaction as a mobile-based service. This also means that SNS can be used as an avoidance space for those with negative urgency, and as an emotional expression space for those with high positive urgency.

• Journal of Intelligence and Information Systems
• /
• v.25 no.2
• /
• pp.99-122
• /
• 2019

According to the recent IDC (International Data Corporation) report, as from 2025, the total volume of data is estimated to reach ten times higher than that of 2016, corresponding to 163 zettabytes. then the main body of generating information is moving more toward corporations than consumers. So-called "the wave of Big-data" is arriving, and the following aftermath affects entire industries and firms, respectively and collectively. Therefore, effective management of vast amounts of data is more important than ever in terms of the firm. However, there have been no previous studies that measure the effects of big data investment, even though there are number of previous studies that quantitatively the effects of IT investment. Therefore, we quantitatively analyze the Big-data investment effects, which assists firm's investment decision making. This study applied the Event Study Methodology, which is based on the efficient market hypothesis as the theoretical basis, to measure the effect of the big data investment of firms on the response of market investors. In addition, five sub-variables were set to analyze this effect in more depth: the contents are firm size classification, industry classification (finance and ICT), investment completion classification, and vendor existence classification. To measure the impact of Big data investment announcements, Data from 91 announcements from 2010 to 2017 were used as data, and the effect of investment was more empirically observed by observing changes in corporate value immediately after the disclosure. This study collected data on Big Data Investment related to Naver 's' News' category, the largest portal site in Korea. In addition, when selecting the target companies, we extracted the disclosures of listed companies in the KOSPI and KOSDAQ market. During the collection process, the search keywords were searched through the keywords 'Big data construction', 'Big data introduction', 'Big data investment', 'Big data order', and 'Big data development'. The results of the empirically proved analysis are as follows. First, we found that the market value of 91 publicly listed firms, who announced Big-data investment, increased by 0.92%. In particular, we can see that the market value of finance firms, non-ICT firms, small-cap firms are significantly increased. This result can be interpreted as the market investors perceive positively the big data investment of the enterprise, allowing market investors to better understand the company's big data investment. Second, statistical demonstration that the market value of financial firms and non - ICT firms increases after Big data investment announcement is proved statistically. Third, this study measured the effect of big data investment by dividing by company size and classified it into the top 30% and the bottom 30% of company size standard (market capitalization) without measuring the median value. To maximize the difference. The analysis showed that the investment effect of small sample companies was greater, and the difference between the two groups was also clear. Fourth, one of the most significant features of this study is that the Big Data Investment announcements are classified and structured according to vendor status. We have shown that the investment effect of a group with vendor involvement (with or without a vendor) is very large, indicating that market investors are very positive about the involvement of big data specialist vendors. Lastly but not least, it is also interesting that market investors are evaluating investment more positively at the time of the Big data Investment announcement, which is scheduled to be built rather than completed. Applying this to the industry, it would be effective for a company to make a disclosure when it decided to invest in big data in terms of increasing the market value. Our study has an academic implication, as prior research looked for the impact of Big-data investment has been nonexistent. This study also has a practical implication in that it can be a practical reference material for business decision makers considering big data investment.

• Journal of Intelligence and Information Systems
• /
• v.25 no.3
• /
• pp.19-41
• /
• 2019

According to the rapidly increasing demand for text data analysis, research and investment in text mining are being actively conducted not only in academia but also in various industries. Text mining is generally conducted in two steps. In the first step, the text of the collected document is tokenized and structured to convert the original document into a computer-readable form. In the second step, tasks such as document classification, clustering, and topic modeling are conducted according to the purpose of analysis. Until recently, text mining-related studies have been focused on the application of the second steps, such as document classification, clustering, and topic modeling. However, with the discovery that the text structuring process substantially influences the quality of the analysis results, various embedding methods have actively been studied to improve the quality of analysis results by preserving the meaning of words and documents in the process of representing text data as vectors. Unlike structured data, which can be directly applied to a variety of operations and traditional analysis techniques, Unstructured text should be preceded by a structuring task that transforms the original document into a form that the computer can understand before analysis. It is called "Embedding" that arbitrary objects are mapped to a specific dimension space while maintaining algebraic properties for structuring the text data. Recently, attempts have been made to embed not only words but also sentences, paragraphs, and entire documents in various aspects. Particularly, with the demand for analysis of document embedding increases rapidly, many algorithms have been developed to support it. Among them, doc2Vec which extends word2Vec and embeds each document into one vector is most widely used. However, the traditional document embedding method represented by doc2Vec generates a vector for each document using the whole corpus included in the document. This causes a limit that the document vector is affected by not only core words but also miscellaneous words. Additionally, the traditional document embedding schemes usually map each document into a single corresponding vector. Therefore, it is difficult to represent a complex document with multiple subjects into a single vector accurately using the traditional approach. In this paper, we propose a new multi-vector document embedding method to overcome these limitations of the traditional document embedding methods. This study targets documents that explicitly separate body content and keywords. In the case of a document without keywords, this method can be applied after extract keywords through various analysis methods. However, since this is not the core subject of the proposed method, we introduce the process of applying the proposed method to documents that predefine keywords in the text. The proposed method consists of (1) Parsing, (2) Word Embedding, (3) Keyword Vector Extraction, (4) Keyword Clustering, and (5) Multiple-Vector Generation. The specific process is as follows. all text in a document is tokenized and each token is represented as a vector having N-dimensional real value through word embedding. After that, to overcome the limitations of the traditional document embedding method that is affected by not only the core word but also the miscellaneous words, vectors corresponding to the keywords of each document are extracted and make up sets of keyword vector for each document. Next, clustering is conducted on a set of keywords for each document to identify multiple subjects included in the document. Finally, a Multi-vector is generated from vectors of keywords constituting each cluster. The experiments for 3.147 academic papers revealed that the single vector-based traditional approach cannot properly map complex documents because of interference among subjects in each vector. With the proposed multi-vector based method, we ascertained that complex documents can be vectorized more accurately by eliminating the interference among subjects.