This study aims to quantitatively visualize the relationship structure between corporate non-financial initiatives and financial performance, and to develop an optimization model that empirically clarifies how different implementation levels of specific non-financial initiatives—namely AI governance (AIG) and Diversity & Inclusion (D&I)—affect financial performance (FP) through the integration of linear prediction terms with a nonlinear evaluation component.

The analysis was performed using descriptive statistics, cluster analysis, correlation analysis, and OLS regression. Building on these results, linear prediction terms were derived using OLS regression and then integrated with a nonlinear evaluation component that captures diminishing returns, thereby formulating the proposed optimization model. Using data from 47 Japanese firms, empirical analysis demonstrated that strengthening AIG had only limited effects on specific FP indicators, whereas reinforcing D&I led to observable improvements. In contrast, a restraint scenario—representing the suppression of non-financial initiatives—resulted in an apparent deterioration of FP.

These findings indicate that the selective implementation or non-implementation of non-financial initiatives (AIG and D&I) significantly influences FP yielding clear managerial and practical implications.

In recent years, recommendation systems have been widely adopted on e-commerce sites, and similar platforms. When user ratings are available, a common approach is to recommend items in descending order of predicted ratings that have been calculated by a model. In particular, it has been shown that utilizing review texts can improve prediction accuracy in such methods. A representative example is Transformational Neural Networks (TransNet). TransNet leverages the ratings and review texts for items that the user has purchased or viewed in the past in order to predict ratings for items that the user has not yet purchased. In this process, the prediction of the ratings is formulated as a regression problem, and the model is trained using loss functions for continuous values, such as squared error. This approach effectively treats ratings as though they are on an interval scale, with equal spacing between rating levels. However, in reality, the ratings only convey the relative order, and there is not actually consistent spacing, making it more appropriate to handle these as an ordinal scale. Motivated by this, the present research proposes a recommendation model that builds on TransNet while reflecting the characteristics of ratings as an ordinal scale. Specifically, rating predictions are formulated as a multi-class classification problem with inclusive relationships among the classes. This allows the model to flexibly learn the intervals between different rating levels, which is expected to improve prediction accuracy. Finally, experiments using multiple real-world datasets demonstrate the effectiveness of the proposed method.

A Low-cost technology capable of identifying the location of evacuees over a wide area is essential for navigating dense smoke conditions during fires in highway tunnels. This study proposes a novel human flow analysis method that addresses this challenge by utilizing Wi-Fi probe requests emitted by smartphones. In order to adapt to the increasing complexity of MAC address randomization for privacy protection, the present method employs six features that are independent of MAC addresses. It applies unsupervised learning, which does not require extensive pre-training, in order to estimate the number of mobile devices, which ensures real-time performance with low-cost equipment.

The present method's effectiveness has been validated through a field experiment simulating evacuation behavior in an actual tunnel. The results demonstrate that by deploying multiple Wi-Fi packet captures at regular intervals, it is possible to visualize time-series changes in the occupant numbers with a high accuracy. Such device count estimation achieved superior accuracy compared to previous RSSI threshold-based methods, particularly for small groups, and successfully visualized the movement of occupants between areas over time. However, the study also identified remaining challenges, including increased estimation errors for large groups, and a tendency to underestimate the actual number of people in real-world environments due to signal shielding by human bodies.

In manufacturing sites, it is important to evaluate the workload of workers on a daily basis, and to implement continuous KAIZEN activities in order to maintain a safe and sustainable labor environment. This study proposes an automatic evaluation method for workloads that is applicable to actual manufacturing sites. Specifically, multiple cameras are used to capture work videos, and the workload is automatically evaluated by classifying the postures of the worker's upper limbs, back, and lower limbs using 2D skeletal data based on OWAS (Ovako Working Posture Analysis System). As a result of applying the method in a logistics factory, it was confirmed that the Action Category (AC), which represents workload in OWAS, could be estimated with an accuracy of 85.9%. However, it was also found that the estimation accuracy of AC4, which indicates the highest workloads, was relatively low. Therefore, the proposed method, which incorporates a human review process in order to supplement the results of the automatic evaluation, is demonstrated to be a practical and useful tool for screening high workloads in actual manufacturing sites.

This study aims to achieve automatic shape classification of cod roe, and automatic detection of Anisakis during the spicy cod roe production process. In this study, a single machine learning model is used to classify cod roe into whole pieces (ippon-mono) and cut pieces (kireko), while simultaneously detecting the presence or absence of Anisakis on the surface of the cod roe. Here, cut pieces are less frequent than whole pieces, and cod roe with attached Anisakis is less common than cod roe without Anisakis. As a result, the dataset exhibits class imbalance. To address this issue, this study applies asymmetric data augmentation using a Generative Adversarial Network (GAN). The results show that the proposed method improves detection accuracy. Furthermore, simulations that incorporate both actual measured values and the accuracy of the machine learning model demonstrate that automation of the detection process can substantially reduce work time.