Search

In the field of retail industry and marketing, identifying customer segments is one of the most important tasks. A meaningful segmentation is able to help the managers to enhance the quality of products and services for the targeting segments. Most of traditional methods used POS data to classify the customer loyalty as “heavy” segment while others are belonging to “light” segment. Based on the previous studies, this paper presents three improvements. Firstly, in addition to customer purchasing behavior, we also include RFID (Radio Frequency IDentification) data, which can accurately represent the consumers' in-store behavior. Secondly, this paper uses broad learning system (BLS) to analyze the consumer segmentation. BLS is one of the most state-of-the-art machine learning techniques, and quite efficient and effective for classification tasks. Thirdly, the customer behavior data used in this paper are collected from a real-world supermarket in Japan. We also consider the customer segmentation as a multi-label classification problem based on both of POS data and RFID data. In the experiment, the results were compared with other popular classification models, such as neural network and support vector machine, and it was found that BLS greatly reduced training time while guaranteeing accuracy.

The scope of this research lies in big data analysis. First of all, the urgency of the problem of not analyzing most of the information that can be recorded and analyzed to improve the functioning of any area of life is stated. Second, the case of a grocery store is used in order to identify useful dependencies. The data from the store contains information that characterizes consumer behavior. As a result, it becomes possible to evaluate the characteristics of interest, such as the average time spent by the buyer in the store, the average speed of movement and movement itself, the distance that the buyer travels during one visit to the store. A linear increasing dependence of the quantity of goods was also found, which, on average, is bought in each department from the inverse average speed of movement of a person in a given department. In the case of the analysis performed, the informativeness and importance of using big data for pursuing profit optimization goals become convincing enough.

In this paper, we propose a purchase behavior model of customers based on shopping path data acquired using radio frequency identification technology in a Japanese supermarket. The proposed model assumes that the in-store customer behavior follows the Markov process. It also assumes that the influence of customer’s stay at the sales floor on customer’s purchase behavior changes depending on their latent state. We employed model analysis to show that customer’s stationary times have a strong positive influence on their purchase behavior at the sales areas such as Western Deli, Japanese Deli, Meat, and Fresh Produce located in the outside area of the store. The stationary time had a negative influence on the choice and weak positive influence on the quantity at the sales areas such as Household Goods, Drinks & Liquor, and Frozen Foods located in the inside area of the store. With reference to this analysis, store managers can set management policies that take into consideration the influence of different stationary times on customer’s purchase behavior depending on the latent state of customers for each sales floor.

Customer relationship management (CRM) is one of the most important recommendation systems to manage customer groups and understand potential relationships. CRM systems use many communication approaches such as telephone, email and social media to collect mass of customer data, which can help the companies (managers) to increase sales growth and enhance customer retention. Since 2000s, the advancements of RFID technology brought a new perspective on customer in-store behavior. Based on such RFID data, this paper describes a novel analytic approach applying social network analysis (SNA) into customer relationship network (CRN) and presents three contributions. First, to be different from existing studies, we attempt to construct the CRN based the shopping path (visiting patterns), which is recognized as a data stream of digitally encoded coherent signals. The CRN can help us to understand the customer relationships from the network perspective. Second, we employ network clustering (also called community detection or node classification) method to extract customer groups by maximizing the modularity. The modularity provides us a structural and relational understanding of customer groups. Third, we use within-module agree and participant coefficient to measure how a customer well-connected to other customers in the CRN; and identify such customers as "hub" customers by a topological process.

Given the prevalence of loyalty programs' implementation in service industries, in order to create a difference in the eyes of the customer from other competitors, examining new loyalty program designs become more and more important for most firms. Compare with Zhang and Breugelmans' research of the item-based loyalty program (IBLP), this research studies a more complicated IBLP design, in which customers can earn different extra points for purchases made on different items. The main purpose of this research is to examine the short-term impact of items with different points in this new IBLP design on different types of customers' purchase behavior. Using data from a Japanese grocery store chain, this study shows that those customers who were heavy customers at the beginning of the IBLP are more affected by this new IBLP design. Then, instead of higher-point items, a middle-level-point, 25-point items has the highest impact on customers' purchase behavior. These findings suggest this special tactic can enhance the value of firm's loyalty program, and help managers to further improve the effect of the IBLP by arranging more targeted items to different types of customers.

In recent years, the consumer tastes have become complicated and store managers are increasingly required to execute multiple promotions and increase sales. In this research, we introduce a dynamic Bayesian model that applies a Cauchy distribution to the estimation of the dynamic state of a product sales trend. The time evolution of the sales data trend reflects customer purchase behavior, and shows a characteristic variation for each brand. We show that the proposed model can better explain the dynamic change of sales by comparison with a regression model. We compare the benchmark and proposed models based on variance, fitness, and prediction accuracy. We show that detecting change-points lurking in time series data of sales leads to important management improvement suggestions.

In this paper, we propose an ASP system for shopping path analysis and a cloud based system with the necessary analysis service for marketing strategies that combines sales data with shopping path data. This paper explains a recommendation system based on position information or an application that calculates a basic marketing indicator using shopping path data by introducing a framework of ASP for shopping path analysis. Because the proposed ASP service is built on a cloud based system, the users can easily access the service through the web based interface and perform large-scale data processing using the computational resources of the cloud based system at low cost.

Over the past decades, prediction of costumers' purchase behavior has been significantly considered, and completely recognized as one of the most significant research topics in consumer behavior researches. While we attempt to measure response of purchase intention to the contextual factors such as customers' age, gender and income, product price and sale promotion, most of business models are basing on a linear equation to estimate weight of these factors due to the linear equation is not only intuitive for other academics to compare and replicate but also luminous to explain the results for business practitioners. Nevertheless, comparing with other research fields (e.g. pattern recognition and text classification), the prediction methods for purchase behavior are overconcentration of the linear models, especially linear discriminant analysis and logistic regression analysis. On the other hand, as more and more information and communication technologies (ICT, e.g. POS and sensor) are introduced into retail, marketing and management to collect business data, the volumes of data are increasing in exponential growth. Analysis based on linear models are insufficient to satisfy the requirement of academics and practitioners any more, and machine learning techniques have been increasingly attracted us to conduct them as an alternative approach for knowledge discovery and data mining. With regard to these issues, this paper employs two representative machine learning methods: Bayes classifier and support vector machine (SVM) and investigates the performance of them with the data in the real world.

The purpose of this research is to construct a sales prediction model for retail stores using the deep learning approach, which has gained significant attention in the rapidly developing field of machine learning in recent years. Using such a model for analysis, an approach to store management could be formulated. The present study uses three years' worth of point-of-sale (POS) data from a retail store to construct a sales prediction model that, given the sales of a particular day, predicts the changes in sales on the following day. As a result, a deep learning model that considers the L1 regularization achieved a sale forecasting accuracy rate of 86%. The products at the retail store have been finely categorized. Even if the attributes of the product categories are increased in number from tens to thousands, the predictive accuracy did not fall by more than about 7%. In contrast, the accuracy decreased by around 13% when the logistic regression model was used. These results indicate that deep learning is highly suitable for constructing models that include multi-attribute variables. The present research demonstrates that deep learning is effective for analyzing the POS data of retail stores.

In recent years, financial crises have occurred frequently in each region, and banks are facing harsh management environments. Bank runs of customers during a financial crisis are one of a bank’s most serious risks. This research aims to build a bank run model for financial crises, use that model to estimate the amount of deposit funds which flow out, and propose a framework for financial crisis risk management. The model proposed in this paper enables understanding of the factors which have the largest impacts on bank runs, providing valuable information for banks to cope with such risks. The model uses survey data, and clarified that bank runs have differences which depend on customer characteristics and branch location. We understood that during a financial crisis, an appropriate branch strategy must be adopted depending on the location and customer characteristics of each bank branch.

The purpose of this research is to develop a framework to analyze the content and a process of persuading process and its application to the communication for the debt-collecting process. It is possible for us to understand how the skilled workers have used the keyword groups concerning the motivation to pay, the payment methods and the payment confirmation in their conversations, to model a persuading process. There is no research and method to deal with a large amount of conversation logs for discovering useful knowledge about a persuading process. In this paper, we were successful in discovering a part of the distinctive features of skilled workers in their conversations for the overdue payment collection, applying our method to communication data in a Japanese telecommunications company.

Decision tree methodology has become an increasingly important tool set in the field of decision science. We develop a multivariate, tree-based decision system for a new application: the determination of whether a newly launched consumer product should be allowed to continue in a highly competitive market. The system is designed to overcome a shortcoming-the inability to capture multivariate interactions-of traditional decision methods. We apply the proposed method to an instant noodle sales data set that contains 38 million transactions, and compare results across several methods.

The advanced investigation of in-store behavior has been dramatically intrigued researchers in the last decade, since the radio frequency identification (RFID) technology was introduced into the grocery stores for information acquisition via tracking the customers. Shopping carts and baskets with a small RFID tag attached are recognized as surrogates instead of enumerators to trail the customers, which can provide an objective and direct observation to measure the in-store behavior of customers. The main purpose of this article is to improve the understanding of purchase behavior of customers with emphasis on meaningful knowledge from the study of their in-store behavior. For this purpose, we present an extraction of purchase behavior using statistical learning theory on dealing with the RFID data, which can describe the purchase behavior into a process of the customers' in-store behavior. In this paper, we introduce three types of factors — personal factor, attitudinal factor and behavioral factor to train a model of purchase behavior, and investigate their effect on purchase behavior among heterogeneous customers grouping by their past purchase incidence. Finally, due to the additional effect of choice switching on purchase intention is also investigated basing on the primary situation, we suppose that these investigation outcomes can result from measuring customers' response to sales promotion due to micro details of RFID data, which can also be applicable to business decision-making implications from the perspective of customers.

The aim of this paper is to propose a new system for the strategic use of customer data that includes and integrates such differing data sources as company databases, mobile telephone networks and Internet data and is a consumer research support system for the discovery of new marketing opportunities. This system, called CODIRO, will be discussed in this paper using a case study of the effects on sales of processed food product television commercials. A system for verifying the validity of consumer behavior models will also be described and discussed. Use of the CODIRO analysis system makes it easy to introduce, into the analytic model, consumer attitude changes and in-store data of many types that have not been used to measure advertising and promotional activity effectiveness in the past.

In strategic management of retail industry, the advanced investigation by using radio frequency identification RFID technology to capture customers' in-store behavior has been dramatically attracted scholars and practitioners in past ten years. As a small RFID tag attached to the shopping carts can be recognized as surrogates instead of enumerators to trail the customers, it can provide us an objective and direct perspective to observe and measure the in-store behavior of customers. In this article, we present a study on this new type of in-store behavior data named RFID data, which can improve the understanding of purchase behavior of customers with emphasis on meaningful knowledge via analysis of RFID data. In contrast to prior studies in this research field, this paper has paid special attention to shopping time that customers spent in supermarket so-called stay time, and presents methodological analysis into two folds. First, we develop a bayesian network BN model to combine both of purchase behavior and in-store behavior as features. As BN is a probabilistic graphical model, it can provide an quantitative analysis process of purchase behavior decision over stay time and also allow us to interpret the decision process of purchasing in a much more intuitive measurement. The results show BN has a better accuracy than other typical prediction models linear discriminant analysis, logistic regression and support vector machine. Second, due to BN can estimate and predict in a nonlinear correlation between purchase intention and stay time, we examine a tedium effect on purchase behavior. During the customers wander in shopping, purchase intention represents a non-monotonic phenomena accounting for the long stay time. Finally, we also investigate the sensitivity and specificity of purchase behavior predicted by our proposal in adjustment of decision threshold and implement several business decision-making implications in actual business situations.

In this paper, we introduce a research project involving the use of various types of data mining technology to analyze Internet Mall Web log data. The objective of this paper is to clarify, using descriptive methods, the process of discovering new knowledge using WEB log data to investigate consumer behavior.

Effective category management by grocery stores requires product category evaluation. Previous studies have evaluated product categories using point-of-sales shopping behavior data. Recent developments in radio frequency identificatio technology facilitate the tracking of customer shopping paths within a store and aggregated stay time in sales areas. In addition to radio frequency identificatio data, we use product discount information from discount flie data and construct a sales area evaluation model incorporating stay time and bargain scale in sales areas. We employ a hierarchical Bayes model to predict the number of purchased products. A typical hierarchical Bayes model considers each customer as a subject; however the proposed model considers each sales area as a subject. From estimation results of the posterior parameter, we evaluate the influenc of stay time and discount scale for each sales area. This provides useful performance measure for the sales areas.

Pharma, a drugstore chain in Japan, has been remarkably successful in the effective use of data mining. From over one tera bytes of sales data accumulated in databases, it has derived much interesting and useful knowledge that in turn has been applied to produce profits. In this paper, we shall explain several interesting cases of knowledge discovery at Pharma. We then discuss the innovative features of the data mining system developed in Pharma that led to meaningful knowledge discovery.

Recently, a wireless non-contact technology named RFID(Radio Frequency Identification) has brought a new perspective on process of purchase decision. Via the RFID tag attached to a shopping cart, the position information of customers in a grocery store can be captured every moment. This paper presents our study based on this type of data. In this study, we transform the RFID data into stay time that the customers spend in supermarket until purchase is decided. In addition to RFID data, a purchase transaction based on POS(Point of Sale) data that only represent the point when customers come to purchase can be extended to a process of in-store behavior. As a probabilistic graphical model named bayesian network is employed to applied, we investigate the stay time how to affect the purchase probability. In the experiment, we also reveal that this affect is different during the purchase decision process by customers in individual age bracket. Moreover, numerical results show our proposal has a better accuracy than other models such as logistic regression analysis.

In grocery stores, discount flyers act as an important tool to provide discount information to customers and spur buying motivation among them. However, discount flyer have limited space. As customers purchase popular products even if they are not actively promoted, mentioning them on discount flyers is an ineffective strategy. Therefore, the proper allocation of discounted products is an important topic which is related to category management. In this paper, we propose a prediction model for the number of purchased products based on the data obtained through radio frequency identification (RFID), point of sales (POS), and discount flyers. We propose a method to determine the share of product categories on discount flyers by evaluating the interaction effect between sales area and bargain scale using the prediction model. The experimental results show that the sales area near the register has a significant interaction effect on bargain sales, suggesting that discounted products should be arranged in front of the register.

Developments in radio frequency identification (RFID) technology have made data on customer movement paths in supermarkets available. In this paper, we propose a method for customer behavior modeling by using RFID data and the hidden Markov model (HMM). In this method, "Stop" and "Pass by" behavior are estimated and the proposed method is evaluated by predicting the sales areas where customers actually purchased items. Using this method, we also demonstrate the shopping momentum. This effect, however, is experienced by only some customers, not all.

This paper introduces an application of the data mining tool, MUSASHI (Mining Utilities and System Architecture for Scalable processing of Historical data), for scientific policy making. Recent advances in information systems have allowed researchers to gather enormous amounts of data on opinions about government policy. However, these gathered data have been individually stored, and have never been integrated because of a lack of techniques to analyze the data in an integrated way. In this paper we propose a cloud application for policymaking during a financial crisis. For this purpose, we have developed an ASP- platform- leveraging distributed- computing technology represented by Cloud computing.

Service science has been raised, and coming to be established as a research domain all over the world. This workshop in Tokyo has been motivated by the systems-design dimension of the service science. We aim to share and discuss a progressive vision to develop methods for innovating systems of service resources where novel values are created and supplied sustainably. A "service system" here is an artificially organized or self-organized active integration of the following resources: Participants, i.e., providers and consumers of services, where a provider of a service may turn into a consumer in different contexts Money, or other entities representing value, and their active flows: An utterance of praise can be also this kind of entity Supply chains i.e., the chain of interactions from creators of service resources (products, information, food, etc) to consumers of services, and Tools (computers, robots, sensing devices, etc) aiding the activities of (1), (2), and (3) In this workshop, we discussed methods for designing and realizing service systems and parts of a service system, with positioning resources of the four kinds above in the systems to be created. By this, we aim to respond to the social demand to design an environment for value-creative and dynamic interactions among participants via resources in the market, rather than merely passing existing products and services from providers to customers for predetermined prices.

Studying shopping behavior is an important and interesting topic for researchers and practitioners. With the improvement of technology in data collection and handling, it is possible and important to take full advantage of these data opportunities to analyze in-store shoppers’ movements so as to understand shopping behavior from different standpoints. In this study, we set up a three-component procedure-based application with the use of a method from Markov chain approach. In this procedure, the sensor network system, the Shopper In-store Movement Graph (SIMG) generator, and the Transition Matrix Compression Algorithm (TMCA) engine, are used to analyze in-store shopping paths and to cluster in-store zones with similar transition behaviors. After clustering, some characteristics of complicated shoppers’ movements can be observed more clearly. An experiment is performed on real data to illustrate that the procedure works in practice. Finally, we close with a brief conclusion and an outlook for the future.

The objective of this paper is to introduce further development of a data mining tool, MUSASHI (Mining Utilities and System Architecture for Scalable processing of HIstorical data), for service computing. Recent advances in information systems have allowed us to gather enormous amounts of data on marketing. However, these gathered data have been individually stored at each company, and have never been integrated because of a lack of techniques to analyze the data in an integrated way and to handle the large amount of data efficiently. To address this issue, we are currently investigating a way to provide a data mining platform as a service so that users can apply various data mining techniques to their marketing data with ease and at a low cost. For this purpose, we have developed an ASP platform leveraging distributed computing technology represented by Cloud computing. This paper describes the ASP platform for data mining services and introduces an empirical application of data mining using our platform.

In this paper we analyze the new type of information, namely RFID (Radio Frequency Identification) data, collected from the experiment in one of the supermarkets in Japan in 2009. This new type of data allows us to capture different aspects of actual in-store behavior of a customer, e. g. the length of her shopping path. The purpose of this paper is to examine more closely the effect of shopping path length on sales volume, which is one of the established ideas in RFID research as well as in retailing industry. In this paper we developed a simple framework, based on criteria of Wandering Degree and Purchase Sensitivity, in order to see how the relationship between distance* walked within the store and sales volume interacts with walking behavior of customers. As a result, in this paper we came up with some useful suggestions for more efficient in-store area management.

Radio Frequency Identification (RFID) technology uses radio waves to track an object to which a small tag is attached. In a Japanese supermarket, we attach the RFID device to the cart and collect data on purchase behavior. In this article, we clarify the relation between purchase probability and the time customers spend in the store section by analyzing the RFID data with main use of descriptive methods. We clarify the way how the stay-time explains the purchase probability and characteristics of each store section. Based on the result, some implications for business are made as well.

Data Mining for Design and Marketing shows how to design and integrate data mining tools into human thinking processes in order to make better business decisions, especially in designing and marketing products and systems. The expert contributors discuss how data mining can identify valuable consumer patterns, which aid marketers and designers in detecting consumers needs. They also explore visualization tools based on the computational methods of data mining. Discourse analysis, chance discovery, knowledge discovery, formal concept analysis, and an adjacency matrix are just some of the novel approaches covered. The book explains how these methods can be applied to website design, the retrieval of scientific articles from a database, personalized e-commerce support tools, and more. Through the techniques of data mining, this book demonstrates how to effectively design business processes and develop competitive products and services. By embracing data mining tools, businesses can better understand the behavior and needs of their customers.

Financial crises are occurring frequently in Asia, Europe, and America, and it is important for banks to investigate strategies for such crises. The objective of this research is to build a model for bank runs by deposit holders in financial crises, and use that model to present a framework for estimating the amounts of deposit withdrawals during financial crises. By carrying out detailed investigation of the bank run model thus constructed, we clarified that the characteristics of customers and branch locations both bring about differences in bank runs. Our estimated amounts of deposit withdrawals during financial crises suggest that each branch should adopt a customer strategy appropriate for the variety of customers of that branch. The bank run model proposed in this research can also be applied to other marketing strategy planning, and has wide applicability.

The aim of this paper is to discuss the development of a system for the discovery of valuable new knowledge and to create effective sales strategies based on that knowledge by using massive amounts of click stream data generated by site visitors. This paper discusses and clarifies the process as to how detailed consumer behavior patterns are extracted from click stream data of Internet mall retail site and how such patterns can be used as a source of new ideas for creating new marketing strategies. We will also discuss our successful use of an improved version of the genome analysis system called E-BONSAI to extract and analyze special character strings related to site visitor behavior indicated by distinctive click patterns.

The knowledge on the relation between a financial state of an enterprise and its future profit will efficiently and securely reduce the negative risk and increase the positive risk on the decision making needed in the management of the enterprise and the investment in stock markets. Generally speaking, the relation is considered to have a highly complicated structure containing the influences from various financial factors characterizing the enterprise. Recent development of data mining techniques has significantly extended the power to model such a complicated relation in accurate and tractable manners. In this study, we assessed the feasibility to model the relation in the framework of data mining, and analyzed the characteristics of the model.

In this research, we discuss the business application of process data, being the massive accumulated time series of changing conditions. Unlike the data resulting from daily operations, process data includes rich information on operational processes. It is unstructured and becomes massive in scale. In this paper, we introduce a project in which we participated, designed to utilize such process data in business. We also examine the lessons learned from this case study.

It is difficult to carry out quantitative measurements of the persuasive power of business communications (i.e., persuasive skills) and such communications are likely to involve difficult to understand, unseen and unknown knowledge. However, using unstructured recorded communication data based on conversations with business customers, we have been developing explicit knowledge concerning skills necessary for effective communications in the form of an expression framework. The objective of this research is to generate a framework and a process for explicit management knowledge concerning understandable communication skills as opposed to the tacit, hard to understand the negotiating skills related to overdue payment collection personnel and to verify the actual usefulness of this knowledge using the accumulated data in a company. Using this process we have developed, it is possible to discover the special characteristics of the communication content of high success overdue payment collection personnel., 2006 IEEE International Conference on Systems, Man and Cybernetics, Taipei, Taiwan, 8-11 October 2006

With the rapid spread of the Internet and the computerization of trading a huge amount of data on the Internet and of transaction database in enterprises has been accumulated. The purpose of this paper is to explain the significance of the technology to process of exabyte-scale data and presents the business application, CODIRO, which will make it possible to integrate various types of large scale data. CODIRO is a consumer research system which discovers new knowledge by integrating the huge amount of different types of data both on the Internet and within companies. This paper will demonstrate the business implications for exabyte-scale information technology research, by explaining an example of the analysis of the sales effectiveness of television commercials using CODIRO., Applications and the Internet Workshops, 2005. Saint Workshops 2005. The 2005 Symposium on 31-04 Jan. 2005

MUSASHI is a set of commands which enables us to efficiently execute various types of data manipulations in a flexible manner, mainly aiming at data processing of huge amount of data required for data mining. Data format which MUSASHI can deal with is either an XML table written in XML or plain text file with table structure. In this paper we shall present a business application system of MUSASHI, called C-MUSASHI, dedicated to CRM oriented systems. Integrating a large amount of customer purchase histories in XML databases with the marketing tools and data mining technology based on MUSASHI, C-MUSASHI offers various basic tools for customer analysis and store management based on which data mining oriented CRM systems can be developed at extremely low cost. We apply C-MUSASHI to supermarkets and drugstores in Japan to discover useful knowledge for their marketing strategy and present possibility to construct useful CRM systems at extremely low cost by introducing MUSASHI.

This chapter describes the framework of knowledge discovery process in sales data and how the data mining system is applied to the data in the real business world by using the domain knowledge. First the framework of the knowledge discovery process in databases is reviewed. It is not clear how users construct the actual data mining process and use the domain knowledge in the existing models. We propose a two-dimensional matrix of knowledge for analysis of sales data to understand the knowledge discovery process from purchase history. Then we distinguish the data mining process from the creation of business action. We point out that efficient knowledge discovery can be achieved by intensively introducing domain knowledge of experts to the creation of business action.

We develop a new method for extracting useful knowledge from individual purchase history of customers by combining information fusion techniques with a data mining tool — string pattern analysis. We demonstrate through several case studies how the method helps firms predict how and when a customer is likely to switch from one brand to another, who is most likely to become loyal to a brand, and when a customer is likely to defect to a competitor. The method comprises several phases of data fusion processes. First, as preprocessing, it transforms purchase history of individual customers into a string of symbols, each of which represents a purchased brand, and then it maps a set of symbols into a much smaller set of new symbols in order to enhance the predictability of the method. Second, as the method produces simple and effective rules described by regular pattern over the set of symbols that effectively distinguish given positive and negative sets of strings. The string pattern analysis retains temporal information in a purchase sequence, an advantage that is not usually enjoyed by other data analytical methods. Finally, we devise a decision tree algorithm and a weighted majority voting algorithm based two-dimensional region rules to aggregate predictors and generate interpretable rules. This is also a data fusion process. The method is illustrated by several applications: brand switching management, customer attrition management, and loyalty program management. The successful applications demonstrate the effective combination of information fusion and string pattern analyses in addressing real business problems.

We have recently developed an E-BONSAI (Extended BONSAI) for discovering useful knowledge from time-series purchase transaction data, developed by improving and adding new features to a machine learning algorithm for analyzing string pattern such amino acid sequence, BONSAI, proposed by Shimozono et al. in 1994. E-BONSAI we developed can create a good decision tree to classify positive and negative data for records whose attributes are either numerical, categorical or string patterns while other methods such as C5.0 and CART cannot deal with string patterns directly. We shall demonstrate advantages of E-BONSAI over existing methods for forecasting future demands by applying the methods to real business data. To demonstrate an advantage of E-BONSAI for business application, it is significant to evaluate it from the two perspectives. The first is the objective and technical perspective such as the prediction accuracy. The second is the management perspective such as the interpreterability to create new business action. Applying the E-BONSAI to forecast how long new products survive in instant noodle market in Japan, we have succeeded in attaining high prediction ability and discovering useful knowledge for domain experts.

This paper presents a new application for discovering useful knowledge from purchase history that can be helpful to create effective marketing strategy, using a machine learning algorithm, BONSAI, proposed by Shimozono et al. in 1994 which was originally developed for analyzing string patterns developed for knowledge discovery from amino acid sequences. In order to adapt BONSAI to our purpose, we translate purchase history of customers into character strings such that each symbol represents a brand purchased by a customer. For our purpose, we extend BONSAI in the following aspects; 1) While original BONSAI generates a decision tree over regular patterns which are limited to sub-strings, we extend it to subsequences. 2) We generate rules which contain not only regular patterns but numerical attributes such as age, the number of visits, profit and etc. 3) We extend regular expression so that we can consider whether a certain pattern occurs in some latter part of the whole string. 4) We implement majority voting based on 1-D and 2-D region rules on top of decision trees.Applying the BONSAI extended in this manner to real customers' purchase history of drugstore chain in Japan, we have succeeded in generating interesting business rules which practitioners have not yet recognized.

This paper proposes a new concept called historybase that helps one to mining data for deriving useful knowledge from a huge amount of data, in particular aiming at business applications. At first we shall point out that data accumulated in databases for daily routine operation is not usually enough for data mining, and that we need to record much more detailed data for the knowledge discovery. Distinguished feature of historybase is to record a stream of events that occur during the course of information processing regardless of the necessity for routine operation, hoping that they are helpful for future data mining. Historybase is already implemented in real business, and has been successfully utilized to improve the business quality

This paper aims to propose models for deposit outflows caused by various financial crises, and to present a framework of knowledge discovery required for bank management to create branch strategies and customer strategies for financial crises. Based on the models proposed in this paper, we understand effects on consumer behaviour caused by various financial crises (computer system failure, scandal, natural disaster, etc.), and estimate the total amount of deposit outflows, based on consumer behaviour data acquired through questionnaires to consumers. A bank can create appropriate branch strategies and customer strategies according to the knowledge obtained from these models.

Consumers classification is one of the most important task in the retail sector. RFID (Radio Frequency IDentification) - A wireless non-contact technology is made easier to classify the consumers’ in-store behavior, recently. This paper presents an extraction of consumer purchasing behavior using statistical learning theory SVM (Support Vector Machine). In this research, we present our recent investigation outcome on the consumers shopping behavior in a Japanese supermarket using RFID data. We observe that it is possible to express the individual difference of consumers how are they spending time (we call it stay time in this paper) on shopping in a certain area of the supermarket. The contribution of this research is in two folds: we employ a SVM model on dealing with the RFID data of the consumer in-store behaviour firstly, as compared with other forecast model such as linear regression analysis and bayesian network, SVM provides a significant improvement in the forecasting accuracy of purchase behaviour (from 81.49% to 88.18%). Secondly, the kernel trick is adopted inside the SVM theory to choose the appropriate kernel for consumer purchasing behavior extraction.

The sensor network technology developed in recent years has made it possible to accurately track the in-store behavior of customers which was previously indeterminable. The information on the in-store behavior of customers obtained by using this technology, namely information on their shopping path, provides us with useful information concerning the customer's purchasing behavior. The purpose of this research is to apply character string analysis techniques to shopping path data so as to analyze customers' in-store behavior, and thereby clarify technical problems with them (the character string analysis techniques) as well as their usability. In this paper we generated character strings on visit patterns to store sections by focusing exclusively on customers stopping by these sections in order to elucidate the visiting patterns of customers who made a large quantity of purchases. In this paper, we were able to discover useful information by using the character string analysis technique EBONSAI, thereby illustrating the usability and usefulness of character string analysis techniques in shopping path analysis.