Ensuring system safety under diverse user operations is crucial in developing interactive systems that must not cause harm. Previous studies explored user modeling based on personality traits and safety verification using model checking, but manual modeling remained a challenge. This paper proposes a semi-automatic method for generating user models to facilitate their systematic derivation. Applied to a railway crossing control case, the method identified safety-critical scenarios associated with personality traits and produced diverse model variations, suggesting its potential effectiveness for safety verification.

Programming contests offer many programming problems that may become helpful materials for beginners. However, those problems are written in “story” style, and the writing style makes the problem similarity evaluation harder and adversely affects an appropriate problem recommendation. This paper focuses on the vector representations of programs instead of the problem descriptions and proposes a novel method for evaluating the problem similarity. Moreover, the paper proposes a new recommendation method combining the similarity evaluation and the solving history and shows its usefulness through an empirical study.

So far, research on code clones involved in microservices has been limited to small-scale investigations. In this study, we conducted an analysis of code clones involved in microservices across 284 projects developed in 12 different programming languages. As a result, we found that code clones written in C# and Java negatively impact maintainability.

There is a known issue in image recognition models where they make predictions based on incorrect or unintended factors. To address this, we explore whether models can be evaluated by comparing the validity of their prediction rationale. Specifically, we propose a method that utilizes LIME to visualize the regions contributing to the prediction, calculates a score based on the consistency of these regions across models, and ranks models based on their total score, with higher-scoring models being considered more reliable. As a preliminary step, we conducted an experiment using the ImageNet evaluation dataset to verify whether models can be assessed by comparing their prediction rationale through visual inspection. The results revealed clear differences between models, confirming that such an evaluation is feasible. Furthermore, we applied our proposed method to 12 pre-trained models using the same dataset. The evaluation results indicate that DenseNet-121 and ResNet-50 achieved the highest scores, suggesting that the proposed method has a certain degree of utility as a simplified alternative to manual (visual) evaluation. In addition, certain models, such as ConvNeXt, exhibited lower scores despite having the highest classification accuracy, as their regions contributing to the prediction differed from those of other models. Despite making correct decisions based on valid reasoning, the proposed method in this paper was unable to distinguish such models from poor ones.

Evaluation of missing data is crucial in assessing quantitative management methods. However, it is difficult to prepare datasets with various missing data for the evaluation of quantitative methods. Therefore, pseudo-missing data are often introduced into datasets. Conventional method for induction missing data often fail to align with missing mechanisms. We proposed a missing data induction method based on missing probabilities. In this paper, we evaluate the proposed method and conventional method based on the perspective of adherence to the missing mechanism. Experimental results show the proposed method adheres more closely to missing mechanisms compared to conventional method.

Although the IFPUG method is the most widely used approach for measuring software size, it requires a longer time to calculate. Therefore, Simple FP and the number of screens have been adopted as simplified size measures. We clarified the accuracy of Simple FP in Japanese software development. Additionally, we demonstrated that early-phase effort is effective in improving effort estimation accuracy based on Simple FP and the number of screens.

Companies are facing the challenge of modernizing software systems that have been running for decades. Incremental modernization, a strategy that improves software systems in a step-by-step manner rather than attempting to simultaneously overhaul the entire system, aims to mitigate the risk of failure. However, this approach can increase costs owing to the complexity of integrating legacy and modernized products. In this paper, we present a case study that employs a cost-benefit estimation analysis in a large-scale industrial project that underwent incremental modernization in the past. We compare the actual and estimated cost-benefit values in the context of incremental modernization. As a result, we confirmed that the cost estimates were valid, but we could not judge whether the benefit estimates were valid.

In this study, we aim to quantitatively clarify the trial-and-error processes involved in error correction activities by novices in Python programming exercises, focusing on their editing content and input operations. The analysis of collected data from programming exercises revealed that trial contents involving input operations required more time and resulted in more extensive edits. Additionally, it was found that during error correction activities, novices temporarily added tokens during the editing process, which were eventually deleted. The proportion of such tokens increased as the number of edits required for error correction increased. These findings quantitatively demonstrate that novices do not correct errors incrementally, and longer correction processes result in a higher proportion of edits unrelated to actual error correction.