In recent years, the proliferation of social networking services has enabled customers to freely express their opinions and complaints. For companies, such review information is important vis-à-vis the evaluation of products and services and establishment of points for improvement. However, systematically extracting useful information from the huge amount of data is challenging, and the current situation entails its subjective interpretation by individual analysts. Herein, using the cosmetics industry as a case study, the objective was to identify customer interests and dissatisfaction quickly and accurately. Specifically, based on review information data, a specialized dictionary for cosmetics was constructed, incorporating vocabulary specific to the cosmetics industry. It was further developed into a generic dictionary for cosmetics that classified negative emotions into “complaints” and “requests.” This made it possible to extract what customers were feeling and complaining about more quantitatively and objectively. While the contents extracted as requests were factors to be improved, it was suggested that information on complaints could also be used to extract factors for improvement. However, it became clear that expressions related to the usability of cosmetics were diverse, and further action is required for review information that includes pictograms and subtle nuances.

This study analyzes how the emergence of generative AI has influenced local government employees’ interest in digital technologies in Japan. Using comparative data collected before and after its emergence, we examine differences across organizational types, job roles, and awareness of work-related issues. The findings reveal that generative AI elicits exceptionally high and uniform interest, regardless of staff attributes, and serves as a catalyst for increasing interest in related, implementable technologies. However, its influence is uneven across technological domains. These results provide new insight into how emerging technologies reshape digital engagement in public administration and suggest strategies for effective digital transformation.

Health foods are subject to diverse regulatory frameworks worldwide, reflecting differences in legal systems, food cultures, and consumer protection approaches. This study compared major regulations in Japan, the United States, the European Union, Australia, and China by focusing on four aspects: definitions, labeling rules, safety evaluations, and enforcement measures. We used the Gemini API, a large-scale language model, to perform a text-based analysis of official regulatory documents obtained from government databases and websites. Our results showed substantial variation in the definitions of “health foods,” with some countries lacking an explicit legal category for these products, whereas others treat them as intermediate products between regular foods and medicines. Labeling rules in all jurisdictions aim to prevent misleading claims and ensure provision of accurate information, although specific labeling requirements differ considerably. Safety evaluations are typically grounded in scientific evidence; however, administrative organizations, review scope, and approval procedures can vary between regions. Enforcement measures for violations also share common elements—such as administrative penalties and product recalls—but differ in severity and legal procedures. These findings demonstrate that health food regulation is shaped by each country’s policy priorities and consumer expectations. The approach used in this study, which integrates large-scale language modeling with comparative legal analysis, can help researchers and policymakers track evolving regulatory trends. The proposed approach can be extended to other regions and incorporate market data and consumer perspectives for a more comprehensive understanding of international regulatory practices.

The omnichannel strategic approach integrates physical and digital channels to deliver seamless customer experiences. While it has been widely adopted in the retail industry, where digital channels play a key role in enhancing consumer loyalty, its effectiveness in manufacturing remains unclear. Because manufacturers face structural barriers in directly engaging with consumers, the consumer experience with digital channels in manufacturing is underexplored in the existing literature. To address this gap, this study investigates whether digital channels affect consumer loyalty in the manufacturing sector. An online survey of 1,500 Japanese consumers across electronics, fashion, and food industries was conducted, and the results were analysed using structural equation modelling. The findings show that digital channels alone do not significantly influence consumer loyalty across industries, age groups, or innovation tendencies ( β = 0.080, p = 0.127). However, for consumers with high product involvement, digital channels significantly enhance loyalty ( β = 0.154, p = 0.045). These findings suggest that targeting young or innovative consumers is insufficient. Rather, increasing product involvement through content marketing focused on product usability and functionality, and combining digital efforts with real-world experiences, is key. This study offers contextual knowledge on when digital channels work, contributing both theoretically and practically to the development of omnichannel strategies for businesses in the manufacturing sector.

This study proposes a novel methodology for extracting technical elements emphasized by corporations in patent documents that uses compound word-based text mining rather than conventional unigram-based approaches. We hypothesize that compound words, which function as bridges between distinct technical elements, play a critical role in patent documents, and we use betweenness centrality to assess their importance. Compound words are extracted by calculating betweenness centrality within the scope of individual claims, using claim-level segmentation based on the structural characteristics unique to patent documents. To mitigate sparsity and bias caused by differences in the length of claim descriptions, compound words with high betweenness centrality are ranked on a per-patent basis, and the frequencies of the top-ranked terms are aggregated by industry and company for analysis. Furthermore, regression analysis incorporating time-series data is used to adjust for short-term fluctuations and develop a mechanism for predicting promising technological elements. The proposed framework is applied to representative Cooperative Patent Classification (CPC) categories in the automotive sector. Leading firms are identified based on the number of patents granted in the target CPC domains, and their technological trajectories are analyzed. Empirical results reveal that these firms not only drive technological advancement within their respective industries but also align their development strategies with overarching trends observed at the CPC category level.

In today’s VUCA era, Companies confront rapidly shifting consumer demands and market landscapes, making new business development indispensable for both sustainable growth and long-term competitiveness. Despite its importance, the failure rate for new ventures remains high, largely stemming from the persistent challenge of accurately discerning market needs and understanding customer dissatisfaction. To address this gap, our study presents a strategy that synergizes large language models (LLMs) with prompt engineering to (1) extract “dissatisfaction entities,” (2) automatically compile a “dissatisfaction dictionary,” and (3) visualize the interrelationships among these detected dissatisfaction factors. Specifically, we employ the Dissatisfaction Survey Dataset furnished by the National Institute of Informatics, narrowing our focus to the “Outdoor Sports_Motorcycles” category. Utilizing a Conversation Chain built atop the OpenAI GPT-4o API and the LangChain library, we systematically process the dataset’s user posts to extract 306 unique dissatisfaction entities. These entities are then distilled into a structured dictionary, capturing the essence of each complaint. Next, we apply TF–IDF and cosine similarity to visualize how these entities relate to one another, thereby exposing clusters or themes of user frustration. Through this analysis, noise and safety concerns emerge as primary, interlinked themes: nighttime engine noise, particularly in urban areas, frequently correlates with perceptions of reckless riding or increased accident risk. These finding highlights both product-design opportunities—such as improved mufflers or noise-suppression technologies—and strategic considerations for municipal policies and traffic regulations. More broadly, our method underscores the power of a systematic, data-driven approach to capturing and organizing customer dissatisfaction, which can serve as a linchpin for minimizing uncertainty in planning and accelerating effective decision-making. By pinpointing precisely where and how customers are dissatisfied, businesses can tailor product enhancements, allocate resources more strategically, and advance new business ventures with heightened clarity.

This study used natural language processing to analyze the content differences between the awarded integrated reports selected by the Government Pension Investment Fund and non-awarded integrated reports. The analysis focused on integrated reports from seven sectors, including consumer-related, basic materials, and processing/assembly categories within the manufacturing industry, and non-manufacturing sectors. Next, latent Dirichlet allocation topic modeling and co-occurrence analysis were employed to examine these reports. The results revealed that the awarded integrated reports tend to emphasize environmental, social, and governance (ESG) elements aligned with industry-specific characteristics. Moreover, such reports are more likely to present a coherent narrative that integrates the resolution of social issues with the company’s own growth strategies. Thus, clearly communicating a value creation story, rather than merely listing ESG-related information, is important. This study provides valuable insights for future integrated reporting practices by highlighting the significance of sector-specific disclosure approaches.

An integrated report is a report that contains information on a company’s ESG activities. For institutional investors who make ESG investments, integrated reports are an important source of information that enables them to learn about a company’s ESG activities. On the other hand, it is not easy for institutional investors to evaluate integrated reports because they are long. Under these circumstances, the objective of this study is to propose a method that identifies characteristics of superior integrated reports using BERT: Bidirectional Encoder Representations from Transformers. To do so, a pre-trained BERT is finetuned to be able to identify texts in superior integrated reports. Then, by looking at the fine-tuned BERT attention, words that contributed to model’s decision can be revealed. To be specific, since these words are labeled with word embeddings, principal component analysis is applied for the word embeddings to calculate the first and second principal component scores for each word. Then, the elbow method is used for the calculated scores to estimate clusters. Finaly, these clusters are considered to represent characteristics of superior integrated reports. This study focuses on integrated reports issued by Japanese listed companies.

With the advent of the centenarian era, the duration of employment in corporations has extended. Consequently, continuous learning and growth, even after assuming managerial positions, have become crucial for individuals and companies amidst changes in the business environment and personal roles. This study attempts to model the awareness of Japanese section and department managers who participated in a three-month, school-style managerial training program. During this program, they engaged in knowledge exchange through discussions with managerial counterparts from other companies. Utilizing text mining analysis of free-text comments from post-training surveys and conducting dependency parsing focusing on frequently occurring verbs, we derived insights. The results indicate that leveraging key terms can elucidate differences in the awareness models of section and department managers, leading to the extraction of valuable knowledge. Furthermore, we propose a method to identify section managers whose awareness aligns more closely with that of department managers by using key terms.

Japanese universities have been focusing on the use of data to improve education. In particular, class evaluation questionnaires in which students answer questions about the classes they have taken are administered at many universities, and as the questions are similar, research is being conducted on how to not only analyze but also apply them to class improvement. Class evaluation questionnaires are administered immediately after classes to determine whether students have acquired knowledge and skills, as well as to obtain evaluations and impressions of the classes. However, it has been pointed out that the analysis of response options is limited to aggregation and that it is difficult to obtain opinions from viewpoints not previously assumed by the teachers. Therefore, in recent years, the analysis of free descriptions is effective in capturing the diverse opinions of students. This study aims to clarify the relationship between qualitative and quantitative evaluations by analyzing numerical evaluations and free descriptions in class questionnaires. The results suggest that it is possible to gain knowledge of the relationship between the answers to the three questions asked in many questionnaires: comprehension of class content, achievement of class goals, and the satisfaction with the class. Supplying this information to teachers, who are the providers of the classes, could be useful to improve their classes efficiently and effectively.

This study presents a method for identifying the skills that local governments need to promote digital transformation (DX) using case information and text generative AI. By using text generative AI, it is possible to abstract the meaning of a prompt based on information and context that is not directly included in the prompt, and form a response based on the broader context and related knowledge. Verifying the reliability and usefulness of output results, this study shows the potential to efficiently identify the skills needed at organizational, departmental, and business unit levels for promoting local government DX.

This study investigates trends in health food research from 1975 to 2024 using text mining and generative AI. Analyzing 92,028 journal entries from Scopus revealed a significant increase in publications, with key topics including nutrients, bioactive compounds, disease prevention, and research methods. AI-assisted classification yielded 12 categories reflecting the field’s multidisciplinary nature. Trends include a growing focus on nutrients, health conditions, study design, and food production and consumption context. The findings highlight the increasingly interdisciplinary landscape of health food research, informing future directions and evidence-based decision-making.

This study explores consumer evaluations of Pacific and Southern Bluefin Tuna parts in Japan, using text mining on Rakuten Ichiba reviews from January 2015 to December 2019. Our analysis identified distinct preferences and concerns regarding texture, fat content, and culinary applications, emphasizing species differences. Texture issues, notably sinews in Southern Bluefin Tuna’s Akami and unexpected bone in its Chu-toro, highlight the need for improved product descriptions and customer education. Findings imply implications for seafood industry strategies, emphasizing accurate product information for enhanced customer satisfaction. Future research should gather demographic data for tailored strategies. This study demonstrates text mining’s role in guiding targeted marketing and refining product in the seafood sector.

This study presents a methodology for quantifying corporate culture based on an analysis of annual reports from manufacturing companies in Japan. The corporate cultural orientation is quantified by analyzing the trends of words appearing in annual reports. Furthermore, the characteristics of their cultural thinking are augmented by comparing the results with the companies’ financial data. The classification of corporate cultural orientations is referenced the “Competing Values Framework,” and we develop a dedicated dictionary for quantifying the cultural orientations. The quantified corporate cultural orientation through this study assists in preventing mismatches in recruitment and considering employees’ career options in Japan’s labor market.

This study determines whether students were able to acquire the desired level of knowledge and skills, by using class evaluation questionnaires which are administered after each class. The questionnaire serves many purposes, but the more questions asked, the lower the response rate. Therefore, this study proposes a method to discover areas for improvement of classes by analyzing free descriptions in questionnaires through students' individual evaluations. Further, we propose an algorithm for discovering improvement objects and deciphering their directions. The proposed method was applied to a case study, and the result suggests that the proposed method is effective.

In recent years, numerous central and local governments around the world have been promoting digital transformation (DX). Although DX is an ambiguous concept, its core initiative involves the introduction and application of digital technology. Unique to recent government efforts is the promotion of utilizing digital technology not only within the ICT department but across a wide range of departments within the organization. The successful use of digital technology depends on the intrinsic motivation of each individual across these diverse department working on the issues in the field. In order to arouse employees' interest in digital technology and encourage them to change their behavior toward its use, it is necessary to understand what factors driving this interest. Accordingly, this research proposes a method to ascertain whether interest in digital technology is affected by its relevance to one's work and one's position within the organization. This method uses "seriousness of consideration" as a scale to measure factors that affect the behavior of local government officials. The method was examined using actual questionnaire data from local government officials, revealing that although general interest in digital technology does not differ significantly among officials, it varies as the seriousness of consideration increases.

To improve product value, research has been conducted on the sound of automobiles. However, studies of perception, including sound, may overestimate the contribution of sound because each perception is treated as an independent object of study. In this study, we clarified the influence of sound on attractiveness in the form of an overarching study of each perception and confirmed the contribution of hearing. A covariance structure analysis of the results of an online survey on the presentation of each perceptual information about automobiles revealed that hearing is the second most important factor for automobile attractiveness after sight.

Integrated report is a report that combines financial and non-financial information and is issued by mainly listed companies. Although it is not mandatory for listed companies to publish integrated report in Japan, number of listed companies issuing an integrated report are increasing every year. It is because creating high quality integrate report is a merit for listed companies, that is, to attract ESG investment from institutional investors. The objective of this study is to discover the descriptive features of integrated reports that are selected as GPIF's “Most-improved Integrated Reports”. To do so, a machine learning based model is proposed with explanatory variables that generated from text data. Also, two specialized dictionaries for integrated report are developed. Then, three models with different explanatory variables are built and compared with AUC-PR. As a result, the model with explanatory variables generated by co-occurrence pairs using two technical dictionaries shows the highest AUC. This result indicates that these explanatory variables are candidates of descriptive features of the awarded integrated reports. Then, by computing Gini importance for each explanatory variable, it reveals that characteristic co-occurrence pairs for the awarded integrated reports. That is, the awarded integrated reports include more statements regarding social contribution and governance and less statements regarding numerical descriptions, compared to the previous integrated reports.

As people enter the 100-year life period, they are working for longer periods of time, and acquiring new skills to stay up-to-date is becoming increasingly important. Moreover, educational institutions for working adults are conducting course satisfaction surveys to improve their educational programs; however, extracting useful information from the comments of several students each year is not easy. Therefore, in this study, a method was analysed to efficiently extract the targets of overall dissatisfaction and demands by conducting reputation and demand analyses using text mining based on the comments of a small group. As a result, we found knowledge that certain words can be used in a tool to efficiently extract relevant information.

Often, existing sentiment analysis determines whether a questionnaire is positive or negative. Unfortunately, this makes it challenging to utilize the negative and correct them. This study used descriptive questionnaires from customers who visited Japanese gastropubs and analyzed them to determine the correct polarity. We used existing sentiment analysis methods to see if the questionnaire was not classified as the correct polarity (positive or negative). Consequently, we found that the whole sentence of the questionnaire was still not always able to judge the content and polarity correctly. Therefore, we conducted an aspect-based sentiment analysis, characterized by a polarity evaluation of certain words in a sentence, rather than a polarity evaluation of the whole sentence. Negative expressions were extracted from the text to determine what was negative. From each expression, it was possible to see what was explicitly unsatisfactory. The extracted expressions were also grouped as aspects belonging to the same group before examining them individually. Subsequently, the extracted expressions were grouped into aspects, resulting in the percentage of each aspect. The results were then used to suggest what could be improved for Japanese gastropubs.

In a society with a declining birthrate and an aging population, infrastructure maintenance is an urgent issue. Properly maintaining infrastructure can curb lifecycle costs, but the inspection process for this purpose is labor-intensive and costly, and is currently not fully implemented by local governments. In recent years, with the development of digital technologies such as AI and IoT, methods to automate inspection work have been rapidly developing. However, their significance and effectiveness have not been fully evaluated. In this study, we will organize the significance of utilizing digital technology in road inspection work, and clarify the trend in its practical application and the effect of its introduction.

Early detection of ambiguities in requirement specifications is effective in reducing the occurrence of specification changes and construction period delays in later processes. However, it is not easy to detect ambiguities in requirement specifications written in natural language without knowledge or experience. In this paper, for detecting ambiguities in requirement specifications, we propose a method of using a knowledge dictionary constructed by focusing on transitive verbs and their objects in requirement specifications and documents that confirm their contents.

Taste is one of the five human senses, but it contains differences and ambiguities among people. Therefore, some researchers consider that taste expression, which cannot be clearly defined, is influenced by the expressions used by others. In this study, we propose a method for quantitatively evaluating user comments on a recipe-sharing website. In evaluating KOKU as an example of a taste expression using the proposed method, we confirmed that expressions from others have a statistically significant influence on KOKU. Although there have been attempts to clarify the influence of other information on taste in cognitive linguistics and biochemistry, these attempts have been based on cognitive linguistics and chemical analysis. The series of analytical methods presented in this paper enables us to quantitatively evaluate the influence of others’ comments on taste expressions the general public has used without clear definitions.

In general, integrated report is a report that combines financial and non-financial information. Unlike annual report, it is not mandatory for companies to publish integrated report. In fact, it is voluntary to do so. However, the number of companies issuing an integrated report are increasing every year, in Japan. It is because better integrate report will attract ESG investment from institutional investors. With number of integrated reports increasing, the Japanese Government Pension Investment Fund, GPIF, awards superior integrated report every year. Its name of the award is called “Excellent Integrated Report”. This award is widely known in Japanese financial industry, but GPIF done not disclose selection criteria that deserve a superior integrated report. In fact, selection criteria are black box. The purpose of this study is to clarify the characteristics of superior integrated reports, by comparing the descriptions in the awarded integrated reports with those in the non-awarded integrated reports based on quantitative methods. The result indicates that while similar subtopics are mentioned in both the awarded and non-awarded integrated reports, the awarded integrated reports mention company's Medium-term business plan which includes topics related to “Corporate value”, “Business growth”, and “Business plan”. In order to be recognized as an “Excellent Integrated Report”, it may be first necessary for the company to create a Medium-term business plan that shows how the company's various activities, including ESG, are related to “Corporate value”, “Business growth”, and “Business plan”.

The global market for bone-conduction headphones is expected to continue to grow, as the COVID-19 pandemic has increased opportunities for telecommunicating, or working from home, as opposed to going to the office every day. Increased teleworking opportunities have led to more online conferencing, and as a result, the market for bone-conduction headphones is growing even more rapidly. Early adopters are very important in today's marketing. They are very sensitive to public trends and tend to adopt new products quickly. Many early adopters gather information on their own, make their own decisions based on the information they gather, and purchase products. And they tend to spread the word about how good the product is and what makes it different from others from a consumer perspective. During a period of market expansion, the diffusion of the word-of-mouth they transmit is key to the diffusion of the product. In this study, we use online word-of-mouth data to study early adopeters’ knowledge of new products.

Under the definition of Baseball Stadium as a specific space, locations from which tweets relating to relevant spaces had been posted were classified whether inside or outside the specific space. As the classification method, BERT being one of the natural language processing models was employed. Through the comparison between the features of tweets inside and outside a specific space, it was revealed that there were differences between them in terms the number of URLs and media including photographs provided in them and shown that classification accuracy would improve by combining the numbers of URLs and media provided in tweets and their contents. In addition, through the extraction and comparison of words affecting the results of classification using LIME, what types of words and information had impacts on the judgement of classification were visualized.

The purpose of this paper is to understand the factors that influence the behavior of addressing conflicts between companies, and focus on comprehending the characteristics of recognizing and addressing conflicts from the perspectives of both the vendor and vendee. For the analysis, a text mining method was used, concentrating on dependency expressions for speech data generated through interviews conducted with a total of 24 people from the vendor and the vendee. The analysis demonstrated that factors such as the interaction with related parties inside and outside the organization, the project being managed, and the importance of the project have an impact on conflict handling.

We study the intended meaning of taste expressions used to describe sake which cannot be clearly defined, by using word-of-mouth information from users on an e-commerce site. Furthermore, we elucidate the taste characteristics perceived by ordinary consumers of sake. We conducted a multiple regression analysis with “Nihonshu-do(Sake meter value),” “San-do(Acidty),” “Amakara-do” and “Noutan-do” as objective variables, and the frequency of various taste expressions, confirmed by word-of-mouth information, as the explanatory variable. Through this analysis, we clarified the relationship between various taste expressions and the “Nihonshu-do,” “San-do,” “Amakara-do,” and “Noutan-do” of sake. We were thus able to make a quantitative evaluation of ambiguous taste expressions related to sake.

The pandemic caused by COVID-19 has also affected the camera industry, and various events have been cancelled. In addition, the recent improvement in the performance of cameras installed in smartphones has reduced the demand for replacement cameras, as it is easy to take pictures without carrying a camera. For customers, online word-of-mouth is what they refer to when purchasing a product. This data is important not only for customers, but also for companies. In this research, we will use online word-of-mouth data and focus not only on numerical data but also on textual data and use text mining to learn knowledge about the decision to replace a camera.

A call center functions to collect and accumulate information. Many enterprises analyze customer voice data and try to use the results for corporate management, such as improving the quality of operator response and handling complaints. In this study, text analysis was conducted on customer–operator conversations at a call center to verify the effect of redundancies on conversations considered effective in achieving customers’ positive evaluation. Specifically, we conducted text analysis on conversations effective in achieving customers’ positive evaluation and those considered not effective. The results showed that redundancies in conversations were related to customers’ positive evaluation.

Open source software (OSS) usage in information systems has been essential. As OSSs are deliverables of the developer community, their quality is not guaranteed. The corporate user adopting the OSS is responsible for its quality. Thus, it is desirable to have information that the corporate could use for quantitatively assessing whether an OSS should be adopted before discussing its detailed contents thoroughly. The objective of this study is to propose an OSS quality indicator model and assessment method by monitoring the status of member raised created and closed issue sessions in the community. The proposed method is based on two axes. First, the trends are summarized in three categories, “Linear,” “Logarithmic curve,” and “Cubic curve,” with thresholds. Next, the deviation timing between the number of created and closed cases is summarized in three stages, “Early,” “Middle,” and “Late.” We attempt to improve the understanding by examining 39 large-scale OSS projects from GitHub. We derived the “T-model” as a quality indicator for OSS adoption decision-making. Assessing only these quantitative evaluations, we can see that the projects in the “Late” horizontal axis and the “Logarithm” vertical axis maintain adequate quality that can be sufficiently used for system development. Based on these results, we conclude that the T-model has shown the possibility of a quick evaluation of whether to adopt the desired OSS for corporate information system.

This paper aims to estimate the population in a specific space from the numbers of posted tweets and their senders, using Twitter’s real-time property and location information data. The population to be estimated was set to be the attendance at each game among the six baseball teams of the Japan Professional Baseball Pacific League held at the main stadium of each team. The relation between the attendance and Twitter data was analyzed, and random forest regression models using Twitter data were used to estimate the attendances. While there are many studies on event detection or location identification using Twitter data, no study has been reported on the estimation of the population in a specific space using “time information” and “location information” characteristic of Twitter data. Using Twitter data, which contains users’ messages, for estimating the population can be extended to various types of analyses, such as the analysis of feelings and opinions of the groups in the space.

Mergers and acquisitions (M&A) help to increase market share and competitivity while come together in the process of transactions are a string of combinations and abortions of duplicated functionalities and divisions, and thus, series of psychological conflicts of employees. This may be very much demotivating as pointed out by critics. It is commonly believed that such conflicts in integrated companies stems mainly from the difference of working cultures. Still, they should be resolved for establishing a new value for the integrated organization. Nevertheless, organizational culture is the sum of all intangible common action pattern of the members, collective norms, governing concepts and beliefs that cannot be easily visualized. It is extremely difficult for the two merging parties to understand and recognize each other’s culture in transactions. Hence, high expectations are on management based on thorough understanding of the merging organizational cultures that could help employees to confront the conflicts and ultimately solve them smoothly. Hosoo and Kimura (2019) have pointed out an easy and qualitative way to measure organizational cultures of integrated organizations through data attained from qualitative interviews.

As citizen report systems are introduced in municipalities worldwide, research on extracting knowledge useful for resolving local issues from accumulated citizen report data is being actively promoted. However, to utilize this knowledge in practice, clarifying the characteristics of the participating reporters and the bias caused by them is necessary. This study focuses on repeat reporters who play an important role and extracts knowledge regarding their behavioral characteristics from citizen report data.

This study evaluates the sentiment of Japanese news and attempts to apply it to investment strategies in individual stocks. When the investment was made at daily, weekly, and monthly frequencies, the effectiveness of the investment was found in daily, but its effectiveness was lost in weekly and monthly frequencies. This reveals that the validity period of sentiment in individual stocks is as short as daily.

Recent consumer behavior research highlights the concern that too many options cause consumers to feel burdened by selection, and hence, reduce their purchasing intention. Existing research focus on the amount of product and information present on product packaging Therefore, in this study, the effects of the number of addition options in the Japanese automobile industry, on the willingness to pay (WTP) were examined, using a randomized controlled trial. Seven categories of options were defined: Color, Interior, Safety, Audio/Navigation, Utility, Technology, and Decoration. The results showed that WTP was not significantly affected by the number of options.

It has been Two years since the bill to legalize casinos in Japan was enacted. Since casinos are avoided by many people in Japan, the promoters try to keep gambling images out of their publicity as much as possible. These promoters are relatively irreverent as to the effect of their promotions in order to enter the Japanese market successfully. Therefore, in this study, the perceptive aspects of those who do not have positive thoughts about casinos have been explored. This was done by designing a hypothetical human cognitive model. Data obtained from a total of 400 subjects through questionnaire survey was analyzed. From the results, it was found that the existing conventional PR methods of promoting casinos as resorts did not necessarily reduce the resistance and aversion to casinos in the minds of their detractors. The initiatives were discussed in order to make the opposition change their minds effectively through educating them to a level of understanding of what casinos are in terms of rules etc., as well as letting them try gambling so as to measure their practical feelings.

In general, product prices tend to fall over time. Also, price drops and changes often take a certain period of time. However, in the EC market, prices do not simply fall, but rise and fall depending on the balance between supply and demand. Furthermore, the frequency of price changes often occurs in hours or minutes. In the EC market, consumers often use a price comparison site to know the market price of a product. On the price comparison site, we can compare a product price of each online store at the same time, enabling to search for the lowest price and the average price of the product. In this paper, by obtaining the information of the lowest price and the average price from a price comparison site and analyzing the relationship between them, we will search for the balance between supply and demand causing the price fluctuation.

The purpose of this study is to evidence a method of understanding the intentions of taste expressions from word-of-mouth data of cooking recipe websites using text mining. This study aims to clarify the use of the word “KOKU” as an example to verify the method. KOKU is one of the taste expressions used like richness experienced in various dishes such as in the taste of wine. In order to clarify the relationship between the features of KOKU and cooking categories, they were clustered using the latent Dirichlet allocation. The categories were classified into groups of foods using similar ingredients, sweetness, oils, and seasonings. Through the analysis mentioned above, the features of KOKU were defined. In the past, there has been no attempt to clarify the features of KOKU using word-of-mouth data from cooking recipe websites. The success in defining “KOKU” is evidence that this method has potential to be extended and applied to expressions other than KOKU.

In recent years, it has become a common function to automatically distribute content suitable for each user by letting AI learn the user’s behavior pattern from the user’s web access log. On the other hand, browsing information by a bot is included in the web access log. There are malicious bots for the purpose of DDos attacks and illegal mass extraction of content. Furthermore, it is not uncommon for bots to disguise themselves as if they were showing their attributes to the user. In this study, we propose a method to discriminate between the user and the bot’s web access log in order to exclude the bot’s web access log from the analysis target.

In recent years, social listening data analysis is adopted in various business areas as a measure to analyze customers’ voices. While sentiment information systems in social listening data analysis have been highly developed, their verification data requires sentiment score by humans. Sentiment means emotions such as Joy, anger and so on. Sentiment score by humans, however, depends on individuals and is not consistent enough. Thus, the accuracy of evaluation is uncertain. If verification data has uncertainty, it is apparent that uncertainty also occurs in the verification of systems. It is therefore important to exclude the uncertainty of verification data as much as possible. In this paper, we studied a standardization method for reducing such uncertainty and its effect. We set three choices, “Positive,” “Negative,” and “Not Sure,” for the verification data used in this study, and sentiment score evaluated the data as social listening data. We first showed an example of uncertainty in the verification data. We then proposed a correction method for reducing uncertainty in the verification data. The method adopted correction formulas using the average sentiment score of all the subjects as the norm. We used these formulas to suppress the bias of sentiment score of each individual subject. Consequently, compared with the result processed by simple decision by a majority, the frequency of “Not Sure” and the amount of mixed data of different sentiment scores decreased. Namely the method was effective in reducing the discarded data of indecisive sentiment score. Furthermore, the variance of data decreased consequently. These results showed that the standardization method was effective in reducing uncertainty in verification data.

Local governments are required to appropriately prioritize and respond to regional issues that are becoming diversified and complicated under the constraints of manpower and budget. In addition, it is required to understand regional issues based on objective data analysis from the viewpoint of “evidence-based policymaking.” Under these circumstances, the “Citizen-Report” mechanism, which has recently been introduced in local governments, may contribute not only to prompt resolution of individual field problems but also to the clarification of the tendencies of problem occurrence. However, the classification of problems is not necessarily set for reflecting the actual occurrence tendencies. Therefore, this study proposes a method to extract and classify problems in an objective and reproducible manner that reflects the tendencies of actual problem occurrences by analyzing the content of the Citizen-Report using text mining. We verify this method using the data of Chiba City as an example, and the result shows that the tendencies of real problem occurrence, which were not able to be understood by classifying based on the category of the department in charge in local government such as “roads” or “parks,” became clear. This method is also applicable to other Citizen-Report data and can be expected to be used for understanding regional issues in various local governments.

In this research, sentiment polarity identification model for finance is developed using financial and economic corpus and deep learning. Specifically, “Japanese Economy Watchers Survey” is used for the corpus and our model accuracy is high. Then the model is applied to evaluate news sentiment for predicting stock return. Our results confirmed that our model captures more news sentiment compared to using common polarity dictionary.

Product price generally tends to decline as time proceeds. However, in E-commerce business (hereafter called “EC”) it sometimes increases influenced by supply and demand conditions; therefore, it is difficult to predict the future price. On the other hand, profit rate does not grossly fluctuate among companies running the same business. Focusing on this point, this study analyzes the tendencies of gross profit using the actual purchase and selling data of a certain EC operator, and also explores for the possibility of estimating the supply and demand of the market by using these tendencies.

In recent years, plagiarism that uses the sentences or phrases of others without permission has become a social problem. It is widely spread from very familiar student reports to novels and worldwide academic papers. In this paper, we deal with plagiarism in student reports, and explain the plagiarism patterns often found there. Then we propose a method to detect them efficiently and accurately. This method is based on the way of making two texts to be compared with into one-dimension string respectively, repeating shift and mutual comparison, and checking the matching section of words. This method can accurately detect perfect matches of any size, regardless of its placement in the text. To this basic detection method, we combine some heuristics which are estimation of detection possibility and compression of strings, to improve both detection accuracy of the plagiarism sections and the reduction of calculation time, and propose it as a plagiarism detection method. This method has only one parameter for plagiarism judgement, and it is also intuitive and easy to set. Finally, we show the effectiveness of the proposed method, especially the introduced heuristics, using real data.

When making a hotel reservation on a private or business trip, users use accommodation reservation websites and use hotel reviews posted on websites as supporting information. While there are a variety of review comments posted on hotel reservation websites, comments differ in content, depending on the difference of users’ purposes, such as a stay in a business area or that in a leisure area. Although there is a reported method that extracts feature representations from review comments in business areas, it is necessary to verify whether feature representations corresponding to each area characteristic can be extracted in accordance with the difference of areas. In the present study, we first graphed impression comments using co-occurrence restrictions and dependency structures employed by the previously reported method and then extracted feature representations by clustering the graph. By applying this method to each of the business area and the leisure area, we extracted the feature representations and the cluster distribution ratios, both of which differ between the two areas. This enabled us to present the order of importance of the decision-making factors for users when they select hotel accommodations in each area.

Open-source software (OSS) has been widely used in the software development process to reduce development cost and development period. However, adopting OSS requires crucial decision-making in terms of various aspects including business, technology, and intellectual property management; these may not be mutually independent and may exhibit a complex set of relationships. This research studies the structured analysis to break down the evaluation criterion axis and the contributing factors when adopting OSS and attempts to clarify the structured evaluation criterion map.

In this research will examine how the news on Japanese listed companies will affect the stock market. Specifically, it clarifies what kind of attribute news influences the stock price of the company from many news information. Our results confirmed that the sentiment that judged the news by the polar dictionary has a certain influence on the stock market. By handling a large amount of information mechanically, investors will be able to conduct more efficient investment behavior.

As IT grows, Web advertisement continues to grow. The cost of Web advertising media in Japan reached 1,220.6 billion yen in 2017. This was 23.6% of the total advertising expenditure. On the other hand, the more involved a product is, the less likely a consumer is to purchase it impulsively due to an advertisement. Those who spontaneously search for information can be expected have similar purchasing behavior. However, there are few examples that quantitatively validate that theory. Therefore, focusing on the Japanese automobile industry, we validated the difference in purchasing behavior (estimation of price on the web) between customers induced by Web advertisement and customers who voluntarily searched for information.

In order to acquire target brand images, companies develop various activities to manage brands as assets. However, in fact, since companies cannot grasp the necessary elements for the image, there are companies that make inconsistent products and promotions into the world. Therefore, in this research, we verified the factors that form "quality" brand image, which is said to be ambiguous and complicated. Quality is said to include not only objective value (Functional value) such as performance and durability but also subjective value (Emotional value) such as beauty and perceived quality. In recent years, companies such as Apple and Samsung, which have excellent emotional value, are emerging, so the value is considered as a major source of competition in the manufacturing industry. We believe that this research will enable companies to undertake effective decision-making without obscuring the elements necessary to acquire the target brand image.

Due to the spread of the Internet and the appearance of smartphones, the size of electronic commerce (EC) market has been growing steadily over the last few years. Despite the tendency that commodity prices generally decline over time, purchase and selling prices in EC business rise and decline according to the supply-demand situation from time to time. This paper investigates the actual purchase and sales data of a certain product sold by an EC business operator and classifies data by using clustering method. Analyzing the results, we extract factors which fluctuate and determine the price in EC market.

In recent years, data on SNS has been gathered and utilized for various marketing activities such as advertisement publicity activities, product planning, etc., are being implemented in many companies. When collecting data, it is common to set conditions such as collection period, language, sending country, and keywords. However, it is often necessary to confront a huge amount of data. Furthermore, it is usual that the collected data contains a large amount of unnecessary noise. Therefore, appropriate classification / extraction work is required in order to reach useful information and hot topics. But the operation is not always easy for everyone; hence we aimed to make everyone easily reach them. This system focuses on "sentiment" that many users are interested in. First, collect data such as posted sentences, extract sentiment (such as good, bad, praise and criticism) in them, and store in the database together with the original ones. This operation is automatically executed. Then, it is surveyed what kind of sentiment is included in a target topic, or conversely, what topic has relationship with a certain sentiment. This system searches information from the previously explained database, and aggregates and visualizes it. This operation is executed based on user’s input. These functions help us to discover hot topics in SNS from various perspectives because the operation is easy for everyone. In this paper, we explain the function, configuration and usage of this developed system.

This paper presents investigating the fraud transaction detection in the mail order industry. These kinds of detection made intensively but the outcome of the research was not shared among the industry. As the B2C industry expands their market size, the fraud transactions increase in number. As a matter of course, this phenomenon is not only continuing but cleverly. One of the conclusive factors for this phenomenon is payment method. That is, the deferred payment method is primarily employed in Japan. The conventional primary indicator for the fraud detection is the ordered time-based information. They are the shipping address, the recipient name, and the payment method. Since conventional detecting method for the fraud depends on some heuristic knowledge, their market size enlargement makes hard to detect fraud transaction. For this background, this paper is presented investigating for comparing algorithms with the actual transaction data gathered from the mail-order industry in Japan. The comparison of weaker learner algorithms is made. The analytical results suggest Random forest is more accurate than XGBoost not only AUC score but parameter tuning costs. This result will make it use for the decision support knowledge for screening customer at the order received phase in the mail order industry.

Due to the spread of Internet, the behavior is changing when customers purchase products. Previously, when customers purchased the products, they asked the experts in the store and got advice. But now, customers think that reviews on the Internet are more important than expert’s advice. Therefore, when purchasing the products, the customers first confirm the evaluation on the Internet. Product evaluation written by customers on the Internet is very important information. And this has an impact on customers’ purchasing awareness. For example, bad evaluation reduces the customer’s desire to choose the product. On the other hand, good evaluation has various of positive effects on products and companies. For companies, it is very important that company’s product strategy is communicated to the costumer correctly. In this study, we compare the results extracted from the score evaluation and word of mouth information with text mining, and suggest a method of the gap between the reputation of reviews and product strategy.

The study of demand forecasting for automobile has been tried for a long time. The demand forecasting for automobile is a quite important matter for business profit point of view. In this report, we show a new forecasting approach to understand the trend of automobile market.

The purpose of this study is to create a model of a relationship in which the dependent variable is the result of a project and the independent variables are the characteristics of human resources. We attempted a comparative evaluation of discriminant analyses with a statistical model and a machine learning model using assessments of the results of projects and team characteristics derived from questionnaire survey data. The results of the evaluation demonstrate that the machine learning model shows a higher discrimination rate within the range of the data used in the analysis, but it became clear that the discrimination rate worsens in comparison with the statistical model when extrapolated.

Hotel reviews posted on accommodation reservation websites are thought to be valuable information for selecting hotel accommodations and also expected to be used for marketing. Since hotel reviews are various in their expressions, it was necessary to make a thesaurus to obtain useful feature representations. Preparing a thesaurus, however, has problems such that it is laborious and requires occasional revisions. In addition, it is necessary to determine subjects of evaluation in advance and set up synonyms for them. Thus, the analysis of subjects not under consideration becomes difficult. In the present study, we first graphed impression comments using co-occurrence restrictions and dependency structures and then extracted feature representations by clustering the graphs. This enabled us to extract feature representations on evaluation from the impression comments in hotel reviews without setting up subjects of evaluation in advance and a thesaurus.

The deferred payment system, which is a traditional Japanese business practice whereby customers do not pay until goods are received, facilitates online fraud. After receiving goods, fraudulent clients simply disappear and the supplier does not receive the payment. However, since the traditional deferred payment system is expected by honest customers, online shopping sites cannot eliminate this payment system, and consequently are vulnerable to this type of fraud. The conventional approach to detect online shopping fraud is the use of various data mining methods based on statistical analysis. In this study, we propose a new approach that does not rely primarily on data mining. The main characteristic of the proposed approach is the use of the nature of economic crimes. In addition, specific implementations to detect online shopping fraud are proposed. This paper explains our approach and also reports the performance of the proposed system in the actual business environment.

In recent years, it has been pointed out that the management competitiveness is shifting from the “functional value” such as high performance and advanced technology to the “emotional value” such as experience and design, but there are few cases that quantitatively demonstrated it. Therefore, we verify the idea that emotional value is more important than functional value based on data of pure recall as a factor that contributes to corporate favor.

Collaborating offline event marketing and online marketing is considered a relatively novel marketing activity for many of the enterprise software companies. In contrast to conventional communication strategies, event marketing features the active and physical participation of limited customers in marketing communication process and online marketing features to expand event awareness to wide range of customers. For both of customers, the companies often have the marketing event to improve software and service awareness in enterprise B2B Software industry. Improving the number of the event attendees is equal to improving the software and service awareness. Therefore, it is required target number of event registration through the website and actual attendance of both new logo customer and existing customers who recently owns or has experiences to use software before. It requires to show the right session to the right customers at the marketing event and website because each user has each different motivation, business scheme, business model, and own role and responsibility. In this study, we could figure out several differences of customer behavior at the marketing event and effectiveness of business impact after the event between 2 segmented customers, such as invited customers and non-invited customers.

In recent years, automatic or semiautomatic processing and analysis systems for a large amount of social listening data have been introduced and used in many industries. Honda Motor Co., Ltd. is also collecting and analyzing voice of customers from much type of media such as SNS using automatic Social Listening System. And verifying whether corporate images and brands are appropriately communicated or not every day. This verification is also used to find symptoms of risk that may be recalled. On the other hand, we found that there were many copied sentences which were delivered from us to society in collected information as voice and opinion of customers. In this case, if these collected sentences are automatically processed as voice of customers using a normal language processing algorithm, we should have a risk to get excessively more positive result than actual. This is because the information delivered from the company like announcement of a new product etc., always includes many positive expressions. And it has been confirmed that the distributor’s advertisement and a large amount of retweets follows it, also causes the same risk for the same reasons. It’s hard to say that we are correctly measuring the voice of customers. Based on the above situations, in this paper, firstly, we named the phenomenon as “Echo Phenomenon”, which incorrectly recognizes delivered information from us as voice of customers. And present it as a problem. Secondly, we propose a simple method to avoid this Echo Phenomenon problem without damaging useful information as much as possible, and show examples of application and its effect.

Customers and IT vendors must be in agreement regarding the process and result of project delivery estimation. However, IT vendors often have difficulty explaining the complexity and dynamic difficulty of project specification for customers, who do not have IT expertise. Moreover, customers may feel unsatisfied because they often receive unexpected project estimations from IT vendors without clarification about the estimation method. This is because the degree of difficulty is not considered in project scale estimations, such as LOC (lines of code) and FP (function point). In this study, we identify potential productivity fluctuation factors that are not considered in software estimation methods such as software LOC, FP, etc. Specifically, the variation factor considered to affect the project is set as an observation variable. Next, the model is evaluated from the population estimation results using the structural equation model. Using this evaluated model, we discover concepts that could not identified before estimation. Additionally, we also identify factors that may affect the composition.

This paper presents investigating fraud transaction detection in the mail order industry. These kinds of detection have done intensively, but the outcome of the research has not shared among the mail-order industry. As the B2C market such as the Amazon type business expands their market volume exponentially, the fraud transactions increase in number. As a matter of course, this phenomenon is not only continuing but clever. One of the conclusive factor for this phenomenon is the payment method. That is, the deferred payment method. The conventional primary indicator for the fraud detection is the ordered time based information. They are shipping address, recipient name, and the payment method. This kind of information makes use of the prediction in common. Conventional detecting method for the fraud depends on the human working experiences so far. From such kind of information, the mail-order company predicts the potential fraud customer with their working experience parameters. As the number of order transaction becomes large, fraud detection becomes difficult. The mail order industry needs something clever detection method. From these backgrounds, we observe the transaction data with the customer attribute information gathered from a mail order company in Japan and characterized the customer with a machine learning method. From the results of the intensive research, potential fraudulent transactions are identified. Intensive research revealed that the classification of the deliberate customer and the careless customer with machine learning.

In recent years, the supermarket industry in Japan is in a state of declining sales over the long term, and market contraction is expected to continue due to environmental changes such as demographic changes. In this research, as a support for supermarket managers placed under such circumstances we suggest a new method to classify good customers who should be kept top priority. We define good customers on the basis of not only current good customers but also customers who generate most of sales in the future and classify good customers in advance from the current information. Additionally We provide goods information for outstanding purchase by good customers to supermarkets. It can be expected to contribute to the improvement of management efficiency by utilizing it for sales promotion activities.

Due to the spread of the Internet, the behavior when customers purchase products is changing. Previously, when customers purchased products, they requested advice from experts in the store. However, now customers think that reviews on the Internet are more important than expert advice. Therefore, when purchasing a product, customers first confirm the evaluation on the Internet. The product evaluation written by customers on the website is very important information. And this is influencing the customer’s purchase consciousness. For example, if a bad evaluation is written for a product. In that case, many customers will not choose that product. On the other hand, a good evaluation has various positive impacts for the products and company. These impacts are very important both for customers and for companies. In many cases, the product’s evaluation page is composing of the score for the predetermined items and personal comment. However, the difference in the score is very small. And it is hard to say the result correctly evaluated the product. Therefore, in this paper, we propose a method to confirm the reliability of score by comparing the result extracted from score and personal communication using text mining and data on the product evaluation page of the camera that can exchange lens.

In software development, there is a need to share a variety of knowledge; therefore, team learning (organizational learning) is required. As tools to support team learning, various groupware has been utilized. In groupware utilization, there is variation among development sites, which is suggested to reflect the maturity of team learning. Therefore, a case analysis on a team with a higher maturity of team learning was performed using groupware utilization as a measure of knowledge sharing. The Gini coefficient is used to represent the distribution of assets in economics. An inversion of the Gini coefficient was used to represent the groupware utilization and defined as the contribution ratio of knowledge. When the contribution ratio of knowledge is large, knowledge sharing is considered to be progressing. The contribution ratio of knowledge in this case study was observed to improve in proportion to the duration of the team. In future, we will expand the measurement range and continue to verify the measurement of team learning maturity using the contribution ratio of knowledge. This study measures the state of the team by analyzing their responses to the questionnaire. If this verification is successful, we would be able to measure the progress of team learning using the contribution ratio of knowledge, which can be measured more easily and objectively without resorting to the questionnaire.

In recent years, the manufacturing industry has seen a shift of the domain of competition from “performance” which can easily be expressed numerically to “design” which is hard to be represented with numerical values. The rise of companies that focus on design, such as Apple, Samsung and IKEA, is remarkable. However, design presents two challenges for the manufacturing industry. Firstly, it is difficult to conduct a questionnaire survey on a design to external customers due to confidentiality. Secondary, subjectivity is involved as evaluators tend to use their experiences and feelings, which makes it quantitative evaluation almost impossible. Therefore, this study takes up automobile exterior design as a subject and aims to realize customer-oriented evaluation process by linking customers’ sensibilities to feature values of automobile exterior designs to enable simulating new designs.

Most of the resources involved in software development are human resources. The software development teams in Japan face various problems. One of these problems is miscasting. When the project managers decide about the staffing of team members, they must understand the collaboration skills of the members for the smooth execution of the project. However, there is no method to measure the collaboration skills of the team members objectively. Thus, there are projects that end in failure because of miscasting. Therefore, it was attempted to measure the collaboration skills and to construct a knowledge base. The knowledge base construction was based on the past interview records. This paper reports the results of a trial in a software development project, in which the collaboration skills of members were discriminated by using interview reports. The interviews with members were conducted by the project manager from the time of project start-up. The records pertinent to the members who had been active as the then current leaders were extracted. These records were used as a knowledge for collaboration skills. This knowledge was defined as Collaboration Skills Knowledge (CS Knowledge). The discriminant analysis in machine learning with CS Knowledge was performed. As a result of the analysis, discrimination rate of collaboration skills was 81%.

To improve own software and service awareness in enterprise B2B Software industry often have the marketing event. Improving the number of the event attendees is equal to improving the software and service awareness. Therefore, it is required to develop the effective web site to promote the event and secure the registration number of event attendance, also asked for cost reduction at the same time from ROI (Return on Investment) perspective. It requires to show the right contents to the right users because each user have several different motivation which is according to their business scheme, business model, and own role and responsibilities. In this study, we could figure out user targeting knowledge through the A/B testing, such as wording effectiveness between the several industries to make the website visitors register the event registration. Also it showed several results of A/B testing to figure out effective schemes and knowledges of user targeting for B2B industry website.

Smart devices such as smartphones mount the wireless communication function by standard features. Although this function is necessary in modern society, the wireless devices sometimes have serious radio quality problems. They include connection impossibly and insufficient transmission speed. It takes long time and much human resources to solve such problems with analyzing each wireless device. Machine learning technologies are utilized in data mining issues and provide possible solutions for several problems. An unsupervised learning method especially is able to reduce the cost of collecting right data and analysis because it doesn’t need the right data. This study proposes the method identifies wireless devices which have radio quality problems by classifying the devices with an unsupervised learning technology from radio information of Wireless LAN. The evaluation experiment on three kinds of unsupervised learning technology with selecting features of radio quality worsen is also described. Gaussian mixture model showed the highest precision accuracy.

This paper presents investigating the customer payment intention prediction in the mail order industry. As the B2C market expands their market volume, the fraud transactions increase in number. The primary indicator for the detection are the shipping address, the recipient name, and the payment method. These information usually make use of the prediction in the Japanese mail order industry. Conventional detecting method for the fraud depends on the human working experiences so far. As the number of transaction becomes large, fraud detection becomes difficult. The mail order industry needs something new method for the detection. The result of the Google Flu Trends shows, accurate prediction needs the heuristics knowledge. For these backgrounds, we observe the transaction data with the customer attribute information gathered from a mail order company in Japan and characterized the customer with machine learning method. From the results of the intensive research, potential fraudulent transactions are identified. Intensive research revealed that the classification of the deliberate customer and the careless customer with machine learning. This result will make use of the customer screening at the time of order received.

A design support system is developed for engineering design based on the ontology which is declarative description conceptualized the world that agents (men and programs) focused on. The design process is understood as determining the design parameters and their relationships which consists the design model. The meta-model of design model is the ontology which represented as a network in the computer system using the XML. The design model is generated from the meta-model as class in Object oriented language . The system built with the above concept provides the following abilities, 1) declarative description that engineers intend to design, 2) effective entering data into Object oriented model, 3)semantic of design model which focused by numerical simulation and graphics programed with Java. Finally, the system's validity and effectiveness is ascertained by applying it to the basic design of an irrigation pipeline.

Although understanding the evaluation and hiring decisions of employers during employment screening is highly important in the sphere of job hunting, the primary factors behind the decision process remain difficult to comprehend. By using the text mining technique to analyze business data from public employment service offices, this report attempts to identify the primary factors behind hiring decisions when the employing corporation is recruiting mid-career candidates. In concrete terms, we first analyzed the results of the employing corporation's hiring decisions and the reasons behind them in individual cases of job referral, as understood by employees at public employment service offices. We then proposed a method to identify the primary factors behind the decision process. Furthermore, we used the proposed method to conduct analyses of the results of the decision process and of rejected applicants in an attempt to understand the primary factors behind negative hiring decisions. The results show that while “experience”, “work”, “age” etc. is extracted, by age groups or occupational categories, characteristic terms are different.

In Japan, general public those who are not medical experts usually buy OTC medicine at a pharmacy, depending on their illness condition. In this case, it is difficult for them to consider how much the OTC medicine is effective for their symptom. The components of OTC medicine have been used as ethical medicines for a long period of time. This is because the efficacy and safety of ethical medicine have been confirmed before being employed as OTC medicine. The information of those confirmed medicines is described in package inserts, which is aimed for medical professionals. Therefore, it is difficult for general public to understand what the package insert describes in terms of medical effects. In this study, from the information which appears in the package inserts of prescription medicines, a method for estimating the effect of OTC medicine is investigated. Also, a method of estimating the effects of medicines without directly compared data is proposed, only by using the information of package inserts of ethical medicines.

It is important to share the reasoning evidence between contractor (IT vendor) and contractee (customer) when entering into a software development project, to achieve satisfactory agreement in estimating development volume. However, IT vendors usually find it difficult to explain the detail of system complexity to customers who have little knowledge about software development. This tends to result in complaint and dissatisfaction for such estimating effort. In this paper, we have applied CoBRA(Cost estimation Benchmarking and Risk Assessment) method for evaluating system requirement, found that the system complexity and the performance requirement are closely related as the cause of estimation discrepancy between IT vendor and customer, and made it clear that the visual explanation for these two factors is a key for the success of software development.

Wireless communication devices such as smartphones are essential in our recent daily life. The devices are also able to apply to understand human behaviors besides connecting a network via high speed communication links. One of such network services includes public wireless LAN services. The service is provided at many public spaces and stores. The stores especially need some added values by the public wireless LAN service. Concrete services such as avoiding disasters and inducement customers via coupons already have begun at many stores in Japan. This study proposes the extraction method of human's standing in line behavior to contribute increasing added values of public wireless LAN services at several stores. The shop owners are able to know how long the standing in line in front of their shops and provide appropriate services to their customers. This study uses wireless LAN frame data on the air. The estimation model with Bayesian estimation method is applied by using the received signal strength in actual wireless LAN frames. It achieved high accuracy of 0.868 through the evaluation experiment by 5-fold cross validation.

This paper presents investigating the customer characteristics of reminder effects in the mail order industry, especially the bad debt customers. These kinds of investigations have not made intensively, performed only such as the shipping address, the recipient name, and the payment method so far and the conventional method for predicting such knowledge depends on the employee's working experiences. For these backgrounds, we observe the transaction data with the bad debt customer information gathered from a mail order company in Japan and characterized the customer with machine learning method. From the results of the analysis, potential fraudulent transactions are identified. Intensive research revealed that the classification of the deliberate customer and the careless customer with machine learning. This result will make use of the revenue expansion with the improvement of the bad debt collections.

Online hotel reservation is widely used for planning travel today. Users of reservation usually refer to the hotel reviews before making their reservations. Numerical evaluation and an impression comment exist in the hotel review. In the case of popular hotels, the number of reviews can become very high, and all numerical evaluations can be fixed at high scores. Therefore, understanding the characteristics of the accommodations is difficult. Numerical evaluation criteria and impression comments are unrelated items. In this study, we analysed the features of the hotel associated with hotel review impression comments and numerical evaluations. We have presented a feature extraction approach for numerical evaluation criteria that are difficult to understand solely from numerical evaluation scores using characteristic expressions of impression comments found in online hotel reviews. And after extracting characteristic expressions from impression comments with text mining, we classified such expressions as positive or negative assessments of the numerical evaluation criteria. As a result, important evaluation criteria can be identified by analysing impression comments from reviews in which numerical evaluation items are scored equally. In addition, by analysing the characteristic expressions extracted from impression comments from reviews in which numerical evaluation of only one item differed, we can identify the reasons for the different scores.

The analysis of text data from social media is hampered by irrelevant noisy data, such as homographs. Noisy data is not usable and makes analysis, such as counting estimates, of the target data diffcult, which adversely affects the quality of the analysis results. We focus on this issue and propose a method to classify homographs that are contained in social media texts (i.e. Twitter) using topic models. We also report the results of an evaluation experiment. In the evaluation experiment, the proposed method showed an accuracy improvement of 8.5% and a reduction of 16.5% in the misidentification rate compared with conventional methods.

This paper presents the evaluation method of the effect range from the package insert of medicine with text mining. Most of the people who take the over the counter medicines cannot understand the medicinal effects. This is because they have little knowledge of the medicine. The ingredients of the over the counter medicines are made from prescription products. The prescription product shows various laboratory findings to obtain authorization from the Ministry of Health, Labour and Welfare. The medical information is described in the package insert of the medicine, and anyone is available. It is possible to evaluate the effect of the ingredient in the over the counter medicines, if we analyse the medical information described in this package insert of the medicine with text mining method. This paper, we focus on both the antipyretic and the antitussive among lots of the over the counter medicines and make intensive research with them. However, variability is observed in the results of the analysis. We analyse the effect of magnification of each experimental configuration. Since the size of this fluctuation range comes from the experimental configuration. Therefore, we summarize the medication type and the experimental data of the medicine in each experimental configuration. As a result, we are succeeding in the effective range of medicine with text mining, if we extract the experimental configuration and analysis above summarization.

Web analytics of B to B sites is mandatory for improving usability and leveraging data for marketing. In this study we tried web analytics by some segmentation and confirmed it is effective. We defined some of the segment models (7 segmentation type) and examined web access using some segments. One of the most important segmentations is registered versus unregistered users and we confirmed user behavior is different with each segment. We confirmed key metrics like bounce rate, referrer, and exit page analysis are especially beneficial for B to B manufacturer site enhancement.

This study proposed the habitual behavior information extraction method from the data on Internet to build effective behavioral change support system so far. It is well known that habitual behavior improvement is important to avoid risk behaviors for a safety driving and a health improvement. It used Latent Dirichlet Allocation approach and evaluated by using telecommunication behaviors in Question and answering Web sites. This paper describes another evaluation by using travel behavior information. On the other hand, the dependency relation is often used to extract valuable information from text data. It also shows the comparative evaluation between our proposed method and the dependency relation method. It is realized the proposed method is more accurate than the dependency relation method according to the result of the evaluation.

This paper presents investigating the customer characteristics of payment method change in the mail order industry. This time we are focusing on the transactional activity of bad debt customers. These kinds of investigations have not made intensively, such as the shipping address, the recipient name, and the payment method so far and the conventional method for predicting such knowledge depends on the employees’ working experiences. For these backgrounds, we observed the transaction data with the bad debt customer information gathered from a mail order company and characterized the customer with machine learning. From the results of the analysis, we are succeeded in characterizing the potential customers. Intensive research revealed that the characteristics of customers who make fraud transactions. This result will make use of the revenue expansion with the improvement of the bad debt collections in the target industry.

The importance of human resource development has been recognized very well; however, it is challenging to realize effective human resource development for many companies especially in the IT service industry. The IT Skill Standards (ITSS) provides indices that identify and systematize business capabilities required for providing IT services. Although the career path model is shown in ITSS, but common skills and knowledge between the job categories and the importance of each skill and knowledge are not shown. Therefore, we extracted common skills between the job categories and computed its score using a Japanese text network analysis.

In general, similar documents searching in the patent field, it is considered effective if compound nouns are used as indexing terms; however, the presence of compound nouns in the patent world is special. Applicants often intentionally create new compound nouns by combining nouns related to their invention that are not in the dictionary. Therefore, compound noun co-occurrence is often rare and document similarity inevitably becomes low. Therefore, it is necessary to find other similar compound nouns. In this paper use the “notification of reasons for refusal”. This is what the examiner to publish. Compound nouns are contrasted for similar inventions from application and citations documents. Extracting these compound nouns, they are then used as knowledge in patent search. Because similar compound nouns are not necessarily semantically related, it is calculated a rating similar to creating a rule.

Information Technology Skill Standards (ITSS) are the indices which identify and systemize practical abilities for IT services. ITSS was published by the Japanese Ministry of Economy, Trade and Industry(METI) in December 2002. Since then, it has come into widespread use as indices of skills of human resources among companies in the IT service industry. Our previous work analyzed the documents of ITSS using data mining techniques, but a proof or evaluation of the results is required. This paper compares between the results of the previous work and the results of a questionnaire that was applied in a Japanese company that uses ITSS to develop its human resources. The questionnaire was applied to 1080 employees in 15 departments in a Japanese company. The purpose of this questionnaire is to grasp the skills of the employees. The correlation coefficient between the questionnaire and the previous work is 0.67 and this is good value. This proves that the results of previous work is very near to the real life.

Generally inquiries through Web forms and e-mails are increasing. These inquiry texts usually include many informal ex- pressions use of the colloquial style and many omitted words. An omitted word causes the meaning of a sentence to become ambiguous and makes the reader misread and misunderstand a context. In this paper we propose a method to predict omitted words from context and knowledge using topic information. From the results of evaluation experiment, we have confirmed that some of our methods can predict omitted words at the accuracy rate more than 40% for the expression that we used in the experiment.

It is important for software testing to create high test case coverage. Test cases are almost created by manually in current situation, so test case coverage is depend on the individual skills. We discuss a method of creating testing pattern for Pair- wise method by using knowledge of parameter values. The method targets functional testing from screens for Web application systems. The method uses knowledge base for identifying pair-wise parameter values by using document analysis to specification documents, boundary analysis and defects analysis, so that it gets rid of dependencies of individual skills. We also discuss case studies which demonstrate the method of creating high test case coverage by using pair-wise parameter values.

Many research studies are being conducted about the analysis of human behavior using sensor devices in the real world, and a variety of information can be found all over Internet. The primary objective is to improve social behavior and habits, such as the prohibition against smoking and the use mobile phones while driving. These unhealthy social behaviors and habits tend to cause health problems and antisocial behaviors. Behavioral modification specialists understand that habitual behavior is one of the most important behaviors in solving these issues. This paper proposes a new method to extract habitual behaviors for discovering the objectives of the behavioral modification. Specifically, Latent Dirichlet Allocation, or LDA, is used for clustering words into appropriate topics of periodical behaviors from literary expressions, and Point- wise Mutual Information, or PMI, is applied to select suitable words for habitual behaviors. The technique by using text data from question-answering websites from the telecommunications industry area was evaluated and showed good performance results.

The online hotel reservation service becomes so popular that the number of transactions has been growing year by year. Before making such reservation, travelers would find it important to refer other guests' opinions about accomodations. In those reviews, both dissatisfied and satisfied impressions used to appear in the same comments. In order to analyze this tendency, we employ text mining and investigate dissatisfied topic from their expressions. We propose some ways to extract useful information from guest review for accommodation's service improvement.

With a rapid increase in size and complexity of software today, the scope of software testing is also expanding. The efficiency of software testing needs to be improved in order to ensure the appropriate delivery deadline and cost of software development. For improving efficiency of software testing, the test needs to be designed in a way that the number of test cases is sufficient and appropriate in quantity. Test analysis is the activity to refine Application Under Test (AUT) into proper size that test design techniques can be applied to. It is for designing the test properly. However, the classification for proper size depends on individual's own judgments. This paper proposes a test analysis method for the black box testing using a test category that is the classification based on fault and AUT knowledge.

IT is a field that requires extensive knowledge. The Japanese government has published the document which defined the knowledge about IT. This document is called ITSS (Information Technology Skill Standard). The ITSS document defines 11 job categories. This paper proposes a method to derive what should be learned for these 11 job categories. In order to learn efficiently, it is indispensable to discern what is important for targeted for learning. By proposed method, important words are extracted from the document of each job categories using cosine similarity algorithm. High weight keywords were used to estimate the most important required education courses for each IT job category.

This paper presents investigating the customercharacteristics from the bad debt list ofa mail order corporation. So far, suchinvestigations have not made intensively, especiallyprivate defaultrisks and conventional method for predicting such risks depend on the employee's working experiences. For these reason, at first,we observedtheactual bad debt list from amail order corporationand analyzedsales data. From the results of the observation, we makeuse of the machine learningmethod to characterizethe potential bad debt customers. Intensive research hasrevealed that the characteristicsof customers, who might fall into the bad debt list, popularitems and so on. This method willmake use forthe revenue expansion;improvement of collectionof the bad debts.

Online accommodation reservations are growing in number, recently. Users of reservation usually refer to the accommodations reviews before making their reservations. However, most of the reviews do not classified for users’ purpose so far and make it hard to understand accommodation reviewing promptly. From those backgrounds, this paper provides analyzing accommodation reviews to find out the characteristics of expressions according to the purpose and examined some methods to present beneficial information at the time of reservation. Moreover, we extracted the differences of accommodation reviews by areas from text mining. From the results of the analysis, we found differences the expressions and the evaluations by area.

A function test for software is one of the Black Box testing methods. As a model for a function test, Decision table is useful to describe logical relationship between operating conditions exhaustively for a single function. However, in case that there are relationships between functions, Decision table cannot describe them. As a result, an insufficient or an excessive testing may be performed. In this paper we expand Decision table model to describe the relationships and propose a design method for function combination testing by using the expanded model. Our method makes test cases which achieve exhaustive combination for functions which has relationships each other. On the other hand, it does not make combination for functions which has no relationship. As a result, we can achieve both exhaustive and efficient function testing.

This paper proposes a method to extract and manage customer preference information used to create My-page, which is a customer service of Internet Service Providers. To provide useful information for customers on My-page, it is essential to accurately grasp the transitions of customer preferences as well as market trends. Customer preference information has conventionally been managed with two-dimensional vectors with customer and preference category axes. In this paper, we propose a method that manages customer preference information with three-dimensional vectors with customer, preference category, and time axes, which enables us to accurately grasp the transitions of customer preferences and market trends. The information volume in three-dimensional vectors is enormous compared with that in two-dimensional vectors. The proposed method addresses this problem by storing only positions and values of the points where information changes over time. This reduces the storage for two-dimensional vectors information per unit time to less than 5% of the original volume. Some kind of knowledge dictionary is required to extract customer preference information. Our proposed method dynamically updates the knowledge dictionary to accomplish a distributed system that performs information extraction at each access point. This enables us to reduce 60% of the CPU usage on the marketing server that is used for managing customer preference information.

Recently, we have been able to obtain information easily from the WWW (World Wide Web). However, it is difficult to find the required information, because the Web space is non-linear space connected by hyperlinks which hides novel useful information. The degree of user interest can be regarded as having an intimate relationship to the number of keywords of the homepage, i.e., meaningful words for the user. This paper describes the possibility of constructing a page-index from the relationship between the user's access record and the information content of the homepage. This information is extracted with respect to the degree of user interest in referring to keywords. We also consider the visualization of Web space. The homepage index constructed is useful for the user's browsing. We have constructed an estimating system and we have evaluated experimental data from several domestic electric product pages. We have confirmed the utility of the page-index and the home page index.

This paper proposes a novel toolkit for educating business people through home-made simulator development. The toolkit is used at our business simulation course. The course consists of (i) Simple gaming experiment among multiple students using Alexander Islands, a tiny business simulator on the WWW; (ii) Lectures to let students to understand the core concepts of systems management through the simulation; and (iii) Home-made simulation model development by the students themselves using the toolkit, which equips a business model description language (BMDL) and a business model development system (BMDS), This paper describes the background and basic principles, the architecture and of BMDL/BMDS, and evaluation results.

In natural language processing, the semantic class information about a word is an important piece of knowledge. For a thesaurus dictionary, which shows the semantic information between words, the editing work is normally carried out manually, which means that a great number of man-hours is necessary for the editing work. This paper proposes a method of extracting the semantic class information on a word from a set of documents. This information is extracted by using the characteristic that the frequency of abstract words is high while the frequency of concrete words is small. As a result of this experiment, it was confirmed that about 20% of the extracted words should be registered in the thesaurus dictionary.