The textile market is depressing, because the market is changing with the short-term changes of customers' demands, whereas the traditional culture old-fashioned companies make trustless decisions on the reasonless feling of staffs touching customers and products. A significant desire, for realizing concrete fruits in this previous situation, was the organizational management of creative ideas on novel opportunities. That is, it often occurs that proposals rising from a certain section become rejected by another section and be killed before its execution even though the proposal might be promising. This is especially troublesome in the case of textile marketing since the marketing decisions should consider subjective impressions of customers about products, which often involves rare occurrences of new demands and preferences. Such rare events can be significant opportunities or risks, i.e., chances, but their uncertainty often disturbs the convergence of the organizational decision of the company. In this paper, we propose a method for chance discovery and its management, customized for textile production and sales company. The method is enforced by visual and touchable data-mining interface of market data, positioned in the organizational process in a company for chance discovery. We show the cases we applied this method for actions in textile-business, evaluating on the performance of data mining methods and the consensus building in the company on the new strategies proposed by the presented method.

View full abstract

The main goal of our research is to establish "methodology for knowledge creation" and to build a supporting system for exhibition planning named "Knowledge Nebula Crystallizer for Exhibition Planning". Every year, exhibition planning companies hold various exhibitions. So far exhibition planning is conducted with implicit knowledge of experienced planners and effectiveness of exhibitions is measured only by questionnaires. In actual situations, it is said that planners cannot obtain adequate and proper knowledge for future planning only from statistical data derived from questionnaires. Then planners fail to evaluate exhibitions they design, and they cannot have accountability enough persuasive for their planning to clients. Planners need to know what visitors to exhibi- tions actually feel and how they behave when they are at the exhibition booth in order to construct strategies for next exhibition planning. In this paper, we are going to describe the methodology adopted for investigating visitors' mental transition in the real world and examples of obtained results. The methodology described here works to articulate exhibition planners' intention and exhibition visitors' mental impression at real exhibition sites, and then to articulate gaps between them. Visitors' interactions with exhibition objects were observed, and their verbal reports (protocol data) and their actions were recorded, collaborating with Dentsu Inc. From the result of our approach, we obtained a prospect that our microscopic approach is useful and effective toward exhibition planning a prototype of "Knowledge Nebula Crystallizer for Exhibition Planning" is described.

View full abstract

This study considered the information presentation method to help the customers make a concept-articulation type of purchase. When customers follow the concept-articulation type of thinking, they only have vague requirements, and try to make a gradual clarification of what they want through the interaction with salesclerks and so forth. We constructed a system called S-Conart (Concept Articulator for Shoppers) to support the concept-articulation type of purchase based on the observations from an analysis of human behavior in actual purchase activities. The user study conducted using S-Conart suggests that changing the contents and/or the method of presenting informa- tion can bring an change to the human mental world, which is also observed when sales-clerk appropriately reacts to the customer in a real-life shopping situation, although in a different form. The result of this user study suggests the possibility of a chance discovery by the customers themselves, which is expected to be useful for building the support system for concept-articulation type of shoppers.

View full abstract

Since there are many resources on WWW, it has been natural that we extract available information from them. In this paper, the system which extracts notable keywords is proposed. Notable keywords are related to user interests and newly appeared words in Web pages. That is, viewpoints and time are important factors. A subject of a topic varies from viewpoints, and the world changes time by time. Thus, notable keywords are useful for estimating future trends because Web pages represent real-time public opinions and user's viewpoint are reflected. Therefore, creative users are able to apply notable extracted keywords to their activities.

View full abstract

Clustering is an important data exploration task in chance discovery as well as in data mining. The first hierarchical clustering dates back to 1951 by K. Florek; since then, there have been numerous algomithms. However, there is no consensus among researchers as to what constitutes a cluster; the choice of the cluster is application- dependent. Although clustering is sometimes evaluated by interpretability of clusters, few studies have been done to reveal the interpretation aspect of clusters. This paper explains development of a new clustering algorithm by graph-based partitioning which aims to simplify interpretation of clusters. Two typical cluster types are considered: a star and a diamond. A star is a cluster with explicit shared context, represented by a central node. A diamond is a cluster with shared context, whose main cause of the context is implicit and hidden. These two types are very easy to understand. We elicit these types of clusters from a given linkage graph. We show some examples and explain the effectiveness of our method.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

In this paper, a method to modify the rule structures of reasoning engines that are composed of if-then production rules is proposed. The modifying structure of rules is effective for multi-purpose of reasoning engines. For this purpose, we use Fuzzy Petri Nets (FPN) because FPN can represent well the structures of rules, and can be easy for modifying their structures. Concretely, we use an FPN to represent a reasoning engine, a learning method from training data on an FPN model, and a method for changing between and/or connectives of rules from the learning results. Finally, a simple example of rule-based decision making systems is given, and usefulness and some future problems of this proposed method are discussed.

View full abstract

In order to design an Adaptive Fuzzy Control System based on the Lyapunov synthesis approach, the fuzzy systems can be viewed as some fuzzy approximators to approximate the unknown functions in the control system. To enhance the quality of the control system, obviously, it needs to tune all parameters in the fuzzy approximators so that the approximation can be improved. On the other hand, fuzzy rules with triangular membership functions in the precedent of the fuzzy rules are used in many industrial control systems. However, how to stably tune the parameters involved in the triangular membership function is still an open problem. The goal of this paper is to design a controller for a class of nonlinear systems using triangular fuzzy functions. The adaptive laws to tune all the parameters in the system are developed. It is shown that the proposed adaptive controller guarantees tracking error, between outputs of the considered system and desired values, to be asymptotically in decay.

View full abstract

Since measuring the electrolyte density is impossible for sealed lead-acid batteries, it is difficult to accurately estimate the residual capacity in any non-standard condition. A popular application like the electric bicycle is therefore problematic because discharge conditions are extremely variable but at the same time an accurate residual capacity estimate is desired. To solve this problem, neural networks were developed to perform this estimation using externally measurable electrical parameters. This is the first neural network implementation to perform this task. It was also found that this solution works reliably even under changing environmental conditions. Moreover, this network solution can estimate the deterioration state of the batteries in just 30s. As a result of this study, a battery checking system using two independent neural networks was developed to estimate the deterioration state and residual capacity of sealed lead-acid batteries in near real-time. This kind of system has large potential in a vast range of battery applications.

View full abstract