This paper implements the infection process of 2019 Novel Coronavirus Diseases(COVID-19) in an agent-based model and compares the effectiveness of multiple infection preven-tion measures. In the model, 1,120 virtual residents agents live in two towns where they commuteto office or school and visiting stores. The model simulates an infection process in which they wereexposed to the risk of transmission of the novel coronavirus. The results of the experiments showedthat individual infection prevention measures (commuting, teleworking, class closing, contact ratereduction, staying at home after fever) alone or partially combined them do not produce signifi-cant effects. On the other hand, if comprehensive measures were taken, it was confirmed that thenumber of deaths, the infection rate, and the number of severe hospitalised patients per day weredecreased significantly at the median and maximum respectively.
ビジネスへの AI 利活用の取組みにおいて,AI 判断の根拠を説明できるホワイトボックスの機械学習手法に注目が集まっている.たとえば,ルールベースの機械学習手法は人間が理解しやすいモデルとして知られている.これらは多数のルールを考慮することで精度を高めることができる.しかし,ルールベースの機械学習はその特徴量として連続値を扱う際には,前処理として離散化してルールの基準を作成する必要があるため,前処理の方法によって精度が大きく左右される.特に,複数の特徴量から形成される高次のルールを扱う予測モデルを構築する場合には,ある1つの特徴量を上手く離散化しても,他の特徴量との関係によって,最適な離散化となっていない可能性が高い.本論文では,ルールベースの機械学習において,最適な離散化を行うための新たな手法を提案する.
News articles have great impacts on asset prices in the financial markets. Many attempts havebeen reported to ascertain how news influences stock prices. However, the limitations in the number ofavailable data sets usually become the hurdle for the model accuracy. In this study, we propose a newsevaluation model utilizing GPT-2. A news evaluation model is a model that evaluates news articlesdistributed to financial markets based on price fluctuation rates and predicts fluctuations in stock prices.Reuter's news texts are classified based on each return through LSTM models. Using co-occurrencenetwork analysis, we reviewed the overview of the news articles retrieved. News articles generated by GPT-2 was used with original news articles, and the model accuracy was examined. The results showed thatcreated news articles are influential over the prediction of stock price fluctuation.
Recently, the form of transaction between consumers called CtoC (Consumer to Consumer) service has expandedthe market scale and has attracted great interest. In CtoC services, where the decrease in users directly causesthe decline of services, it is necessary to take measures to prevent existing users from leaving. In order to supportsuch measures, the problem of predicting user departure with high accuracy and interpretability has been activelystudied as "User Churn Prediction". The purpose of this research is to construct a high-performing and interpretableframework to predict user settlement and churn that incorporates not only users' own characteristics but also theeffects of contact between users. We applied Graph Neural Networks(GNNs), which have been attracting attentionin recent years, to the task of predicting user settlement and churn in the CtoC service.
In order to realize EV (Electric Vehicle) society, the development of EV charge infrastructuresis indispensable. Under current situation, existing EV charge infrastructures are insufficient. If the numberof EVs continue to increase with this pace, there will be long queues at EV charging stands, and the waitingtime for charging EVs in cities will become much longer. However, if we naively increase capitalinvestment, the low profitability and investment return risk would be a problem as well. In this study, weaim to develop a smart EV charging reservation system that combines a dynamic pricing system accordingto the degree of congestion and a car navigation system that recommends and reserves EV charge standsthat match user's preference (time or price). Using this smart EV charging reservation system, we candisperse their charging locations and decrease total waiting time for charging. We will evaluate the effectof this new system by means of multi-agent simulation.
This paper provides the technological structure transition of FPGA through the patent analyseswith making graphs using classification codes in patent information. Knowledge extraction from patentinformation has been made so far, but conventional patent analysis methods which depend on personalheuristic knowledge make it hard to extract the technological structure. We are focusing on the classificationcodes in patents that are assigned to define the patent's technological fields. With the proposed method, wewere succeeded in extracting the technological structure transition of FPGA.
It is said that staff engaged in nursing care and related services have high job stress and highturnover. The purpose of this study is to clarify what interventions are needed to increase job satisfactionof caregivers and prevent them from quitting job or switching companies. Surveys were conducted ofworkers providing nursing care and related services at two companies, both of which are located inHiroshima Prefecture and have the same executive head. The data from the employees' stress check as wellas the data from the employee interview are used for the analysis. We used machine learning like decisiontree and random forest models to identify important factors. This scheme can be applied to other cases.
The purpose of this study is to propose a new risk assessment method that complements thecurrent audit risk assessment being conducted by experienced auditors. Specifically, we consider a methodof discriminating high-risk financial figures in account units from the ones to be audited through machinelearning using public financial information. From the practical view point of financial statement audits, therisk of account balances being audited is often recognized by the change in balance compared to the sameperiod of last year and the change in correlation with other related account balances. We use, therefore, theMahalanobis distance as the evaluation variable, a distance concept considering the correlation betweenvariables. In conclusion, we propose a method of discriminating high-risk account combinations byclustering, an unsupervised learning method in machine learning.
Corporate valuation is an important process for determining corporate value for managersconsidering initial public offering (hereinafter IPO). However, in the IPO, the underprice phenomenonwhere the public price falls below the initial value has occurred. Due to this phenomenon, new publiccompanies suffer opportunity loss. In this study, we consider the cause of this phenomenon to be in theevaluation of corporate value and try to improve it by using patents in existing methods. Specifically,SCDVs are created based on patent documents held by companies and applied to similar companycomparison guns to evaluate the validity of the open price.
In today's complex and uncertain business environment, traditional method of decision-makingand organizational practices are no longer enough viable and require significant change. It is same for thedesign of products and services. Mindsets and methods that is suitable for the new era are required in thedesign process. Then we have developed an educational game to learn and discuss such mindsets andmethods. Our game has a ruleset assuming the simple system design and other ruleset assuming the complexsystem design. By playing this game, students can understand the problems caused by complexity and howto deal with them.
We propose a mapping model for describing customer experience transition processes inbusiness innovation cases: System-Experience Boundaries Map (SEBM). Different from the othercustomer experience mapping methods, SEBM focuses on potential boundaries, which an innovatedsystem restricts customer experiences. SEBM represents a customer side process of business innovationas a resolution of those restrictions. Using together with the managerial decision-making descriptionmodel we previously presented, SEBM describes value co-creation processes in actual business cases. Wealso consider SEBM application to facilitation and log-analysis on business gaming.
Information security incidents could be one of the most concerns for corporate activities.Considering these factors, this study attempts to clarify the relationship between information securityincidents and stock market reactions. In this analysis, we analyze the stock market in Japan from March2005 to January 2015. As a result of intensive analyses, we confirm the following results; (1) informationsecurity incidents have a negative impact on corporate values, (2) information security incidents withsignificant impact have long-term negative effects on stock market.
In this study, we developed a lead qualification method to run effective lead nurturing campaignsfor shopping complexes and EC sites with various shops. By using this method, it is possible to extractpromising prospective customers and understand the characteristics of customers who use or do not usestores. We found that the model improves the accuracy by incorporating the customer's data about otherstore usages. Our method would help shop managers analyze the characteristics of the customers who usethe store.
本研究は、飲食店のPOSデータと顧客データを用いて企業の経営力向上に繋げるシステムの提供を目的としている。昨今のビッグデータ活用時代において、個人のデータを蓄積し利活用することの重要性が増してきている。個人が頻繁に利用しデータ取得可能な業態として飲食店が挙げられ、そこで蓄積されたデータを分析することでCRMに有用な示唆を得ることができると考える。本分析では、顧客の購買行動の特徴を明らかにし、在庫管理と販売政策について検討を行った。
In this study, we investigate urban structures that affect urban parks.Urban parks are important places for urban life, safety, and comfortable cities. However, urban parksoften cause problems. We have not found a solution to these problems yet. In order to solve theseproblems, it is necessary to understand the situation related to complex urban parks. And it is important tobe able to verify the impact of the situation on urban parks. This time, we will clarify the urban structuresurrounding urban parks and consider the effects of urban structure on urban parks. This study has thepurpose of realization of a comfortable urban park. This study is in progress.
本研究では、日本の株式市場における Twitter のツイート情報と小型株の株価との関連性について分析を行った。機関投資家と個人投資家の投資手法の違いを踏まえた上で、GoogleCloud Natural Language API を用いた感情分析を行い、ツイートと小型株の株価の関係を明らかにする。
When predicting stock prices with a complex model using machine learning or artificialintelligence, overfitting sometimes occurs, and the prediction accuracy expected in actual operationcannot be obtained. In such a model, the cost function is presumed to be steep and multi-modal, while in amodel that maintains stable prediction results, the cost function is considered to be gradual andsingle-peaked. In this study, we first compared the performance of several stock price prediction models,and then visualized the cost function for each model using t-SNE. As a result, the model using Lassoregression, which had the highest performance, showed a gradual unimodal cost function, while the linearregression, which had relatively low performance, showed a steep and multi-modal shape.Visualizing the cost function using t-SNE can be an important index for evaluating the stability andversatility of a stock price prediction model.