This study aims to build a support method for consulting service companies allowing them to respond to client demands regardless of the expertise of the consultants. With an emphasis on the revitalization of small and medium-sized enterprises, the importance of support systems for consulting services for small and medium-sized enterprises, which support solving problems that are difficult to deal with by an enterprise, are increasing. Consulting companies can respond to a wide range of management consultations; however, because the contents of a consultation are widely and highly specialized, a service proposal and the problem detection depend on the experience and intuition of the consultant, and thus a stable service may occasionally not be provided. Therefore, a support system for providing stable services independent of the ability of consultants is desired. In this research, as the first step in constructing a support system, an analysis of customer information describing the content of a consultation with the client companies is conducted to predict the occurrence of future problems. Text data such as the consultant’s visitation history, consultation content by e-mail, and call center content are used in the analysis because the contents explain not only the current problems but also possibly contain future problems. This research proposed method for analyzing the text data by employing text mining. In the proposed method, by combining a correspondence analysis with a DEA (Data Envelopment Analysis) discriminant analysis, words that are strongly related to problem detection are extracted from a large number of words obtained from text data, and variables of the DEA discriminant analysis are reduced and analyzed. This paper describes improved method for the application in the real problem. The method is improved to eliminate the following two problems. First, IDF values are used to extract more general phrases. Second, in order to reduce the number of companies that cannot be identified, it is used standardization and data are expanded with synonym dictionaries. In this study, computer experiments were conducted to verify the effectiveness of the improved method through a comparison with an existing method. The results of the verification experiment are as follows. First, there is a possibility of discovering new factors that cannot be determined from the intuition and experience of the consultant regarding the target problem. Second, through a comparison with the existing method, the effectiveness of the proposed method was confirmed.

With the development of IoT techniques, it become easier to collect users’ action data. By analyzing and using those data, consumers and producers will mutually exchange their intelligence and better customize product development processes. This study focuses on the users’ daily life, examines a proposed system using sensor shoes with several sensor devices embedded in the insoles, collecting action data of users, extracting their action features, and then issuing some advice to help users train more efficiently. As described herein, a service model uses a backpropagation (BP) network to distinguish users' actions and to extract their action features using Self-organization Map from the presented sensing data. Experiments to confirm the feasibility of the proposed methods are undertaken. With the former result indicating an overall accuracy of 89.61% in distinction of 5 actions: sit, stand, walk, run and jump. The latter results showing that SOM is helpful to classify action feature in detail. After the analysing parameters of each cluster numbers, it is possible to make the action feature visualized by providing a colored cluster map, which makes it easier to compare train periods. Results also show the potential of utilization of data collected by devices to provide personal service.

In our previous paper, we showed the M-tree index [7] using GiST in the PostgreSQL database. In this paper, we present that result and we extend that with some preliminary experimental results with other indexes. We compare the M-tree index with the BK-tree and the VP-tree indexes. These can be work in metric space with edit distance, that can be used to compare DNA sequences or melody of songs. In this paper, we compare the indexes in PostgreSQL. We use the range based queries to analyze the performance of the indexes. The result shows that the M-tree index is faster than the other two indexes.

In recent years many frameworks have appeared, which enable users to easily build, visualize and execute deep learning networks on graphical interfaces. However, they do not always provide enough opporunities to automate this process. Generally, data processing programs can be organized into dataflow graphs that define the operations to be performed sequentially on the data. The operation of deep learning neural networks can also be interpreted in a similar way, in which the input data to be processed is a specific data set and the operations to be performed on the data are the layers of the net. Due to architectural reasons, the entire deep learning neural network graph must be built before actual running, thus it is necessary to change topological execution of dataflows to evaluation preceding graph building since knowing the layers separately is not enough to operate the nets. As a solution for displaying editable program graphs, we created a framework in which data processing related to Python packages can be described and the programs built from them can be visualized and executed (mostly) automatically.

Reducing the number of features when applying machine learning algorithms may be beneficial not only from the standpoint of computational cost but also of overall quality. Wrapper-based procedures are widely utilised to achieve this. The choice of the wrapper is of utmost importance. Bio-inspired computation algorithms represent a viable choice and are widely adopted. Due to the sheer number of available algorithms, this choice could prove to be somewhat difficult, especially since not all are made equally. The aim of this paper is to explore several optimisers on diverse datasets representing classification problems in order to evaluate their performance and suitability for the task of feature selection.

With an ever-increasing number of connected devices, parts of the decision process necessary for IoT environments are shifted from the cloud back to the local network. The onset of so-called Edge computing increases the demand for intelligent modular gateways. These gateways should be able to support a large number of different sensors, actuators, protocols, and applications. In this work, such a modular intelligent gateway for decentralized IoT environments is designed. The implementation of the gateway is based on Windows 10 IoT Core and Raspberry Pi 3. This gateway allows to plug-in sensors, protocols, and data processing applications in real-time, without the need for gateway restart.