Published papers

There has been a notable trend of high social interest in the impact of artificial intelligence (AI) on research and development (R&D). While there is much discussion regarding the productivity improvement effects that AI brings to R&D, most of these discussions are qualitative in nature, and studies that quantitatively verify these effects at the corporate level are limited. In this study, we vectorized all Japanese patents from 2024 using Sentence-BERT and conducted analysis through t-SNE visualization targeting multiple IPC classifications. The analysis results revealed that (1) vector representations using large language models have a certain correlation with IPC classifications, and (2) through evaluation using this analytical method, there is potential for more precise patent classification than conventional IPC classification.

This paper describes a constructive architecture for using large language model (LLM) in generating virtual personas with respect to the authors' proposed simulation-based quantitative persona creation method. This architecture consists of a layered structure for persona generation and the concept of operational evolution. The layered structure consists of a Fundamental Layer consisting of typified facts extracted from simulations, a Narrative Layer generated from facts typified by LLM, and a Presentation Layer that uses LLM to enhance the realism of the personas as personas. In the operational phase, this layered structure allows for easy modification and reflection based on feedback from the field and changes in the environment. The main findings of this study are as follow: our architecture could 1) achieve logical consistency with simulation and survey results, 2) generate stories with a certain paragraph structure, 3) inject external information and knowledge, and 4) derive the technical process naturally to generate the personas. In particular, we conduct specific demonstrations about 3) above mentioned; we found that it is possible to revise persona stories in a realistic manner by adding external information to them through prompt engineering methods like Reasoning and Action; ReAct.

本研究は，人工知能に関連する特許の可視化を試みたものである．分析対象は，日本企業が出願した特許とし，分析対象期間は2024年とした．本分析においては，特許データのベクトル表現を取得した後，可視化を行った．ベクトル表現の獲得には，Sentence-BERT等の複数のモデルを採用した．本分析では，各モデルの特徴などの知見の獲得および得られた結果の応用について検討した．

本研究は、同一の経済イベントに対する株価反応時間の差異に着目し、米国、日本、台湾の三市場における情報伝播構造を比較分析することを目的とする。金融市場の連結性が高まる中、同一情報ショックは複数市場へ波及するが、価格反応の速度や調整過程は一致しない。その差異は、市場流動性、投資家構成、産業構造、情報処理行動の違いを反映している可能性がある。本研究では、2000年以降の主要株価指数を用い、イベント後の価格調整を時系列的に測定することで市場別の情報遅延を定量化し、その比較から各市場の経済特性と意思決定様式を検討する。

Persona methods are widely used in marketing to create customer-centric services. However, the following issues exist in designing and evaluating personas for services and measures that have characteristics which may benefit customers in the distant future, such as in the domains of asset formation and life planning. First, many analyses are based on previous or current customer attributes, which do not always result in accurate customer classification. Second, identifying the appropriate target group is difficult because the impact of the services and measures provided to each persona is not evident. In this paper, we present an example of persona creation and evaluation that considers the above characteristics, targeting the asset formation domain. This method generates future customers attributes using a social simulation based on real data, understands the effects of measures using scenario analysis in a virtual space, and has aspects as a proposal for improving quantitative persona creation methods. The key findings are that predicting and grasping the possibility of future change in customer attributes and creating personas based on these predictions is quantitatively and objectively possible; moreover, quantitatively grasping the impact of services and policies on customers and prioritizing among the created personas is possible.

We used text mining techniques on review sites where employees can post evaluations about their own company. By sentiment analysis, we measured the working environment score from the review data. We examined how this score affects the company's performance. The result showed a positive correlation between the score of the working environment and the operating profit margin. Furthermore, by conducting morphological analysis, co-occurrence keywords, and dependency parsing, we were able to identify and visualize common words between companies.

社会シミュレーション分野では，リアルスケールのシミュレーション実施に向けた基 本技術が整備されつつあり，「デジタル社会実験」の実現についても議論がなされている.金融 分野でも，社会シミュレーション技法は様々な場面で活用されている.資産形成やライフプラン ニングに関し，個人の属性に即した解決策提示や施策検討などの諸課題に対しては，当該技法を 用いたアプローチが有効となる.本稿では，著者らの最近の退職前後世代の資産形成・取り崩し のシミュレーション;仮想的施策の影響度分析;定量的エクスペリエンス・マップ作成手法;顧 客とサービス提供者の共創的状況の形式記述などの研究成果に基づき，ライフプランニング領 域におけるデジタル社会実験の実現に向けた考察を行う.

The study acquires sensory evaluations of property floor plans as a form of tacit knowledge of experts and examines the impact of these evaluations on the price of condominium unit (Unit). The floor plans of Units are evaluated by the expert sensory with their ease of use and efficiency, but it has not been quantified to date. This study targets the floor plan images of the second-hand Units for sale, and these sensory evaluations acquired from experts. The multiple regression analysis for the Unit price is conducted by adding the sensory evaluation of the floor plan to the candidate factor variables such as building age and total exclusive area. The result of the multiple regression analysis indicates that the sensory evaluation of the floor plan was selected with the statistically significant level in the model. Therefore, it was suggested that the sensory evaluation of the floor plan has a certain impact on the Unit price.

We conducted text mining on data from review sites where information and reputations about companies can be shared, and evaluated the labor environment score from that data through sentiment analysis. As a result of investigating the impact of the labor environment score on corporate performance, a positive correlation was shown between the labor environment score and the operating profit margin. Furthermore, by conducting morphological analysis of the review data and analyzing co-occurring keywords, we were able to identify common words between companies and visually confirm them.

This study uses a large-scale language model to examine methods for preventing frailty. In Japan, one out of every 2.5 persons will be 65 or older in the first half of life in 2060. Compared to healthy elderly, frail elderly people are at about twice the risk of needing assistance or nursing care. This study examines prompts to prevent frailty, improve quality of life, and realize a sustainable society. Utilizing a large-scale language model, we confirmed the feasibility of several prompts and the challenges.

As Japan’s labor force becomes increasingly reliant on foreign workers, linguistic barriers have emerged as a major challenge in workplace safety. This study explores the use of large language models (LLMs) to generate “Easy Japanese” from safety manuals in the manufacturing and agriculture sectors. Focusing on prompt design, we compare Zero-shot and One-shot input formats using GPT-4o and a specialized Easy Japanese model. We evaluate output using three methods: manual evaluation, automatic evaluation, and machine-based self-evaluation using ChatGPT. The results show that both input styles produce usable outputs, with One-shot offering greater accuracy and Zero-shot offering speed and ease of use. These findings highlight the potential of LLMs to support low-cost, multilingual workplace communication and promote multicultural inclusion.

ニュースは金融市場の資産価格に大きな影響を与える．ニュースと株価変動の関係性を分析し，ニュースを評価する取り組みはこれまでに多く行われており，ニュースと株価変動の間には関連性があると報告されている．しかしながら，ニュースは非構造化データであり，定型的に精度高く処理し，分析に用いることは難しい．本研究では，多くの自然言語読解タスクでSOTA(State of the Art)を達成している BERT を用いて株式変動を説明するニュース評価モデルを構築し，実証分析を行い，従来の分類モデルよりも高い正答率を得られるか検証を行った．本研究における分析の結果，BERT を用いたニュース評価モデルの精度が最も高かった．

Business cases include a lot of information on corporate activities, including informa- tion contained in securities reports. In this study, we analyze the relationship between the orga- nizational structure of Japanese IT companies and financial performance using business cases. .. The organizational structure of Japanese IT companies can be broadly divided into pyramid-type organizations and flat-type organizations. Recently, IT companies have been called for flatten- ing, and many companies are moving to flat-type organizations. However, Japan is originally a pyramid-type organization, and it is thought that there are not many companies that are emerging as flat-type organizations. The purpose of this paper is to show a concept for exploring one of the factors.

We use an agent-based model to elucidate the strategies of firms that have achieved superior performance even in an environment of market shakeout due to innovation. Assuming that a company’s sales consist only of intangible assets, we analyze the specific actions of the company regarding investment strategy from the viewpoint of decision-making and found that (1) it is important to invest more than competitors even at the expense of efficiency in the early stage of a business, and (2) business growth by investing talented human resources ahead of competitors is the most important factor in the (3) If it is difficult to secure human resources, it is important to make investment decisions for further new business development ahead of competitors. One of the novelties of this study is that it quantitatively provides valid suggestions regarding investment strategies for businesses.

In this research, we used an agent-based model (ABM) to forecast the spread of electric vehicles (EVs), one of the next-generation vehicles. We constructed two models to predict the spread of EVs: a consumer model that uses general consumers who own private cars as agents, and a corporate model that uses companies that own vehicles for business purposes as agents. Ultimately, the predictions of these models are used as one of the necessary data to predict the future demand for gasoline and diesel oil, and the results are reflected in the gasoline and light oil distribution model.