論文

This research focuses on the decline in demand due to the spread of next-generation EV and hybrid vehicles, uses an agent-based model to predict gasoline demand, and predicts demand for a delivery model in which gasoline is delivered from a refinery or oil depot to a gas station by tank truck. We conducted a search for efficient delivery methods by reflecting the prediction results. The following results were obtained. Although oil depots have the function of reducing the number of tank trucks needed, using oil depots tends to increase delivery costs. If future gasoline demand decreases, the number of tanker trucks needed will also decrease. This is expected to reduce the need for the depot's ability to reduce the number of tanker trucks needed, making closure more economically advantageous.

電気自動車(以下、EV)など次世代自動車普及の方策を立案する目的としてエージェ ントベースモデル(以下、ABM)を用いた研究は多数なされているが、一般消費者への普及予測を 対象としたものが多く、企業への普及予測を行なったものは限定的である。本研究では、消費者 モデルに加え、企業モデルの構築も行なった。企業への普及予測も行なうことで、今後のガソリ ン・軽油の需要に与える影響をより正確に把握することが期待される。EV などの次世代自動車の普及予測を行ない、ガソリン・軽油の配送モデル構築を行なうことを本研究の最終目的とする。

In recent years, information issued by companies has become more diversified, increasing corporate transparency to stakeholders. Especially, in CEO messages, which are a source of information indicating a company's future vision, changes over time are likely to emerge. Therefore, we collect CEO messages for each year as text data, vectorize them using natural language processing techniques, and then calculate the cosine similarity between messages for each year. Cosine similarity is a quantitative measure of similarity between documents by calculating the cosine of the angle between two vectors. Using this method, we quantitatively evaluate the similarities and differences between the CEO messages of each company for each year to identify trends in the real estate industry and the differences and characteristics of each company. As a result, it was observed that when social conditions undergo significant changes, the CEOs work to further promote the traditional businesses held by the companies by adapting the direction of the companies to the social conditions.

顧客中心のサービス設計のため，マーケティング領域では，ペルソナ手法が広く用い られている.デザイナーのインタビューによる定性的な作成手法に加え，昨今では，多量のデー タと様々な分析手法を組み合わせた定量的ペルソナデザイン手法が多用されている.本稿では，著者らが提案しているシミュレーションベースの定量的手法について，ペルソナの企画者やデザイナーなどの実務有識者にヒアリングを行う.当該手法の有用性や限界を議論するとともに，今後の改善に向けた方向性を議論する.

As a case study of the Shibuya and Kanda Station area, a correlation analysis was conducted using building and location indicators including the visibility and the integration value based on the space syntax theory as factor variables and the capitalization rate as an objective variable. The multiple regression analysis was then conducted. The results showed that the integration value as the space syntax theory indicators had an impact on the capitalization rate.

This study will identify the superior model by grade recognition of building images of office buildings and residential buildings by two image recognition models that are known to be highly accurate. With respect to expert grade evaluation in facade images of buildings, for practical use in the real estate business, focus on the layer depth of convolutional neural networks (CNN), an AI image recognition technology, and build two models using VGG-16 and ResNet-50, which are different layer depths. Modify the network so that it can be applied to the model, perform transfer learning, and obtain a transition graph of classification accuracy. The results of this accuracy comparison considered showed that VGG-16 outputs higher accuracy than ResNet-50 for both applications, indicating that VGG-16 has the advantage. This also suggests that the performance required of the model (architecture) is not only based on layer thickness, but also includes the need to avoid overlearning due to layer depth.

Product-harm crises occurred worldwide, and the effects are reflected not only on customers but also on enterprise value. In our study, we analyzed specific sites which publish scandals in Japanese and identified the keywords of product-harm crises using co-occurrence analysis. Based on the results, we can analyze news data to understand the effects of product-harm crises via event study. We will also analyze the news in different countries by natural language processing and aim at finding a difference in product-harm crisis management.

Crypto asset (Virtual currency) transactions have long been in the limelight. Crypto assets use cryptography to achieve decentralized management by participants on the Internet. Since they can be used by an unspecified number of people and there is no intermediary in the transaction, they are often misused for money laundering and terrorist financing. In this study, we adapt various machine learning methods to ELLIPTIC DATA, which is BITCOIN transaction data used in many previous studies. As a first step, we examine parameter optimization of random forests, feature refinement, sampling methods, and so on...

2060年、日本では高齢化率が39.9%およそ2.5人に1人が後期高齢者（65歳以上）となる。本研究では、健康寿命の延伸に向けてフレイル予防のプロンプト構築と自然言語処理技術（NLP）を活用したプロンプト評価を行った。とくにフレイル予防や健康支援の分野では、EBMに加えユーザーに個別化されたアプローチが効果的である。本分析では、複数の大規模言語モデル（LLM）による比較を行うことでモデルの特徴と課題を確認した。

本研究では、会話の継続性に焦点を当てて、大規模言語モデルを取り込んだフレイル予防の方法を検討する。フレイルをもつ高齢者は要支援・要介護になるリスクが約 2 倍になると知られている。個人の人生の質を左右するフレイルに対して予防的に取り組む重要性は高い。本分析では大規模言語モデル(LLM)を活用し、複数のプロンプトから実現可能性と課題を確認した。

In this research, we will create a model that estimates the attributes of followers based on the information of utterances on SNS as clues about what kind of customers support e-sports teams and their sponsor companies, and compare the attributes. The purpose is to clarify the sponsor's integrity by The survey targets e-sports teams that operate Twitter accounts and their sponsor companies. Using the information posted on Twitter, we will compare the attributes of followers between target accounts and analyze the trend of followers. Based on the analysis results, we propose indicators for optimal matching of teams and sponsors.

本研究は、顧客の時間感度が異なる状況における配送戦略と顧客満足度の関係を分析した。分析の結果、特定の顧客を優先しすぎることで、全体の経済効率性を損なう可能性があることを見出した。UAVと車両のハイブリッド配送システム研究に関し、興味深い結果を示したものである。

Purpose – This study examines trends in the foreign ownership ratio of Japanese real estate investment trusts (REITs) from 2014 to 2023. Using panel regressions, it explores how firm characteristics, macroeconomic factors and policy interventions shape foreign investment patterns, offering insights for managers, investors and policymakers. Design/methodology/approach – Using panel regression (fixed and random effects) on 33 J-REITs, this study analyses firm-level, asset type, sponsorship and macro-financial factors as well as the impact of inclusion in the FTSE EPRA/NAREIT Global Index effect on foreign ownership. Findings – Market capitalization, yen appreciation, hotel sector exposure and global index inclusion are positively associated with foreign ownership, whereas higher leverage, Bank of Japan J-REIT purchases, stronger ROA, higher policy rates and logistics sector exposure are negatively associated. Practical implications – The findings provide actionable insights: for investors, market size, sector and index inclusion signal liquidity and accessibility; for J-REITs, asset risk–return characteristics, sector choice and leverage discipline matter; for policymakers, index engagement and monetary policy influence foreign capital flows. Originality/value – To the best of our knowledge, this is the first study to apply panel regression to foreign ownership of J-REITs, highlighting sectoral and macro-financial drivers and providing evidence from the world's third-largest REIT market.

Recently financial education has become more important because financial technology is highly developing and financial market is growing in importance. In this research, we apply Business Game method to financial education. Especially we focused on learning of asset allocation. As a result of intensive experiments, we found that (1) players learned not to take excessive risk through business game and (2) they recognize the importance of risk control by our experiments. These findings indicate that our approach is valid for financial education.

In this study, we construct an analysis system for the impact of mergers and acquisitions (M&A) on innovation activities using large-scale patent data in Japan, the U.S., and German. In the analysis, we focus on the technological similarity between the acquirer and the target companies in M&A. Specifically, we use natural language processing to analyze the patent document data and measure the technological distance among companies. In addition, we measure the innovation output in M&A using the patent data and classify the analysis targets according to the size of the technological distance. We also analyze the trend of the innovation output after the announcement. By conducting the analysis, we confirm that the innovation output after M&A announcement tends to increase in the M&A group with medium technological distance. In this result, one of the novelties is the applicability of unstructured data and machine learning methods to the research field of M&A and innovation.

In the field of marketing, 'persona' is widely used as a method to support customer-oriented product and service design. However, in many cases, the persona design process is confined to the classification of customers based on limited attribute information projected from the past to the present using cross-sectional data. In this study, we propose a method, based on attribute information extended by social simulation, to improve the persona design process. The framework consists of the following procedures: 1) treat changes in the external environment explicitly and handle target customers' possible attribute changes (i.e., 'virtual life'); 2) classify customers based on a large quantity of attribute information supported and extended by social simulation. As a demonstration of the proposed method, this paper presents a specific case study that focuses on the customer asset situations. We conduct a simulation of customer asset situations at a future point in time, taking into account asset succession and price fluctuations of risky assets, based on individual questionnaire data concerning asset formation and withdrawal in retirement. We design personas using the conventional method and the proposed method, and we compare the two. The results show that the proposed method facilitates a) estimation of future changes in customer states, b) incorporation of possible future events in each persona, and c) consideration of future customer state changes in the state of classification of personas.

Physical AIの進展により、サービスロボットの実運用導入が拡大しているが、配置戦略の経済性を定量的に評価する枠組みは十分に確立されていない。本研究では、POSデータに基づき、需要、人的生産性、ロボット補完効果、および運用コストを統合したロボット配置戦略の経済性評価フレームワークを構想する。さらに今後は、人間–ロボット協働ポリシーを導入し、利益最大化配置点、配置可能領域、およびDeployment Frontierの観点から、配置戦略の経済性検証の拡張を検討する。

Various means exist for obtaining information about a company. However, many of them are not timely and, from the investor' s point of view, do not have enough information. In particular, it is even more difficult to find out trends of diversified companies with multiple segments. Therefore, this study aims to extract timely information on the trends of each segment of a company from news data using BERTopic. The analysis targets news headlines of diversified electronics firms listed on the Japanese stock market. The sample period was 24 years, from 1996 to 2019, and the number of news items for analysis was 26,058. As a result of the analysis, we found that (1) BERTopic can classify the target news into 46 topics, (2) it is possible to identify company segments and extract trends in company activities from the classified topics, and (3) it is also possible to visualize the time-series transition of topics related to each segment. (4) The results obtained from the analysis were used to determine the value of the company's investment in the market. In addition, the results obtained from the analysis were consistent with the descriptions in the annual reports. These results indicate the possibility of obtaining highly immediate information on corporate trends, such as R&D, through the analysis of news headlines via BERTopic.