Background music (BGM) is one of the key elements that constitute digital games. BGM plays an important role in entertaining players and shaping the game’s atmosphere, and high-quality BGM can enhance the overall evaluation of a game. In game development, there are cases where creators compose the BGM themselves instead of outsourcing it to external composers. However, composition often requires knowledge of music theory and experience, making it difficult for anyone to easily create BGM. In this study, we trained an LSTM model to learn melodies of BGM based on the impressions of maps used in role-playing games (RPGs). Using the generated melodies as references, we then composed BGM for use in the same RPG. Furthermore, we analyzed the generated results and training data and conducted comparisons with learned and generated melodies using other methods.

　This study proposes and evaluates a method to support readers’ emotional understanding of novels before reading. Based on Plutchik’s eight basic emotions, 8-dimensional emotion vectors were assigned to both novels and music, and music was selected for each novel using cosine similarity. Emotion vectors for 56 novels were generated via BERT-based sentiment analysis of reviews, while 80 piano pieces were quantified through a user survey. In an experiment with 20 participants, cover images, synopses, reviews, and other related information of unread novels were presented under two conditions ― with and without music ― and participants predicted their emotional responses. The average cosine similarity between participants’ response vectors and the proposed method’s novel emotion vectors was statistically significantly higher in the music condition than in the no-music condition. These findings demonstrate that presenting music based on emotion similarity effectively supports readers’ emotional understanding of novels.

　In this study, we propose a song recommendation method that combines acoustic compatibility with novelty based on inter-genre distance to address the ”filter bubble” problem in music subscription services. By consolidating 181 genres into ”macro-genres” and normalizing differences in statistical variance, we create serendipitous discoveries while preserving user preferences. The results of the present evaluation experiments showed that the novelty metric is effective to a certain extent, achieving a score of 0.80. However, a linear novelty model that simply integrates different scores overemphasizes novelty due to discrepancies in score variance, leading to the inclusion of heterogeneous genres that deviate significantly from users’ inherent musical preferences in the top recommendations. This revealed a key issue: approaches that simply maximize the distance between genres or use unnormalized integration methods present ”noise”(useless heterogeneity) that undermines user preferences, rather than beneficial ”serendipity”.

　A skill tree is a character development system in which players start at a designated point and, by spending points earned within the game, progress across a grid—much like a board game—to acquire skills (abilities). Skill trees are a form of character development that adds a new dimension of enjoyment to games and are used in a wide variety of titles. In traditional games, many skill trees are created and fine-tuned manually by developers, and each title features meticulously crafted level design. However, there is a challenge in that their design and adjustment require a significant amount of time and effort. In this study, we analyzed the structure and skills of skill trees implemented in existing role-playing games (RPGs) and based on these findings, developed a prototype system for the automatic generation of skill trees in role-playing games.

This paper reports on the methodology and outcomes of constructing a TEI XML corpus of early Coptic manuscripts, focusing on the Gospel of Judas (Codex Tchacos). We designed an encoding schema based on the TEI P5 Guidelines that accommodates the morphosyntactic characteristics of the Sahidic dialect, with a particular focus on verb-argument structure annotation. Our tagset enables precise description of transitivity distinctions, direct object marking patterns, and structural alternations involving nominal and infinitival complements. The corpus maintains interoperability with Coptic SCRIPTORIUM while providing a foundation for systematic retrieval and quantitative analysis of verb-object co-occurrence patterns, including theological vocabulary unique to Gnostic texts. We present the design principles, encoding examples, and prospects for future linguistic and stylometric analyses.

　In this study, we developed a literature search support tool based on survey papers to assist novice researchers in their literature searches, and verified its effectiveness. This tool visualizes the structure of a research field by focusing on the chapter structure and reference relationships within survey papers. Additionally, it displays a list of keywords and references for each field. Results from an evaluation experiment involving 11 undergraduate students in their first through third years showed that 10 respondents reported being able to grasp the overall structure of the field, indicating that this tool has the potential to support a comprehensive understanding of research fields. On the other hand, it was also confirmed that 9 respondents were unable to sufficiently recognize the keywords. This is thought to be due to issues with visibility or task instructions. Moving forward, it will be necessary to consider improving the presentation of information regarding the overlap between fields, visually emphasizing keywords, and integrating the tool with search behavior.

　We comparatively analyzed external link references in two Japanese online community question answering (CQA) services, Yahoo! Chiebukuro and Oshiete! goo. Using large-scale datasets, we examined basic characteristics of external link use, frequently referenced websites, and referenced YouTube videos. The results showed common major sources such as YouTube and Japanese Wikipedia, as well as clear platform differences. References to past questions within the same service were prominent in Yahoo! Chiebukuro, whereas references to the National Tax Agency website and other CQA services were characteristic of Oshiete! goo. The content of frequently referenced YouTube videos also differed markedly between the two platforms.

　Characters are one of the most important elements in a narrative, and among them, heroines are often given particular emphasis. Introducing a variety of heroines can be considered an effective way to enhance a story’s appeal. However, when the number of heroines becomes too large, there is a risk that the overall impression of each character may be diluted, making it necessary to present multiple heroines as simultaneously appealing. Previous studies have analyzed factors that enhance the appeal of individual characters. However, it remains unclear which attributes and relationships among multiple characters contribute to making them appealing at the same time. Therefore, this study aims to clarify methods for presenting multiple heroines as simultaneously appealing by conducting a quantitative analysis of heroine attributes in works featuring multiple heroines. Specifically, multiple regression analysis, factor analysis, and cluster analysis were employed. The results revealed that factors such as emotions toward the protagonist and relationships with other heroines strongly influence their appeal.

　This study analyzes the decision-making process of title creation using generative AI, focusing on a poster production task within a computer science course for university students. In addition to the final documents, students submitted comprehensive worksheets recording their AI interaction logs, selection processes, and justifications for their choices. The analysis identified a four-stage deepening process of editorial judgment. It revealed that proactive learners intervened in AI’s suggestions by specifying target audiences, supplementing usage contexts, and imposing formal or emotional constraints. This demonstrated a process of converging AI outputs into a range of "appropriateness" envisioned by the students. These findings suggest that learners do not merely accept AI proposals but prioritize "editorial judgment" to adapt content to specific social and cultural contexts. This report details the mechanisms of student decision-making and the diverse context specifications provided. Furthermore, it discusses the necessity of evaluating the "process" of AI interaction in education and explores the establishment of new assessment frameworks, including the essential role of faculty feedback in fostering information literacy.

At the National Institute of Informatics (NII)，more than 100 procurement projects involving external vendors are handled annually. In this process，the development of precise requirements and the preparation and review of procurement specifications impose a significant workload. In addition， issues such as incomplete or ambiguous requirements，inconsistent formats，and the tacit nature of expertise adversely affect both the quality of specifications and operational efficiency. This study aims to address these challenges by proposing a framework in which generative AI collaborates with humans to interactively refine requirements. The study also explores the automation of procurement specification creation and review processes. We developed a prototype system in which AI assists in checking draft procurement specifications，and evaluated its performance. Through a comparison with human annotations，we assess the effectiveness of AI-generated feedback，and further examine the impact of incorporating historical review logs on improving accuracy. Finally，we discuss the effectiveness of the proposed approach and outline future directions toward automation.

　This study presents an approach to structured extraction of historical knowledge from Latin inscriptions using LLMs. A data model covering person-/ communities, their careers, benefactions, and social relationships was designed based on epigraphic expertise. Structured JSON data was extracted from 7,597 inscriptions across 12 cities in Roman North Africa, validated through a two-stage verification process, and converted into 314,456 RDF triples using a custom ontology. This enables new possibilities including content-based search and analysis.

　Demographic and Health Surveys (DHS) are large-scale household surveys conducted in low-and middle-income countries, providing open-access, individual-level data on a wide range of health indicators, including maternal and child health and nutrition. However, their recognition remains limited outside the field of public health. In this study, we used the Tanzania DHS (2022) to estimate the prevalence of stunting, underweight, and overweight among children under five years of age at the regional level. Furthermore, we constructed a composite index integrating undernutrition and overnutrition, and visualized the spatial distribution of the double burden of malnutrition (DBM) using a geographic information system (GIS). The results revealed substantial regional disparities in DBM, with particularly high co-occurrence of undernutrition and overnutrition observed in regions such as Njombe. With the global promotion of open data and the increasing availability of open-source software such as QGIS, this study illustrates how public health analyses and policy recommendations addressing regional issues in different countries can be made more widely accessible to the public.

　While the public release of educational digital archives and various LOD datasets has progressed, mechanisms for federated search and utilization of these resources based on the Course of Study have not yet been sufficiently developed, posing challenges for their effective use in educational settings. This study constructs an educational resource LOD infrastructure centered on the Course of Study LOD by linking Textbook LOD, Unit LOD, video contents from NHK for School, and S × UKILAM Teaching Material LOD, and proposes a federated search method across learning materials that integrates and presents related teaching materials and video contents starting from textbook units. In addition, the search results are ranked based on the hierarchical structure of Course of Study codes so that educational resources with higher relevance to unit content are prioritized. A search experiment targeting a unit in lower secondary school social studies history confirmed high relevance among the top-ranked results. These findings suggest that the proposed method is effective, to a certain extent, for discovering educational resources related to a given unit through federated search.

　To date, quantitative studies on narrative structure have been conducted across various genres. However, there has been very little quantitative research on the narrative structure of comedy. Furthermore, while several omnibus style animations have achieved long-term popularity, quantitative characteristics have not been extracted until now. Therefore, this study aimed to clarify the characteristics of omnibus-style comedy anime by focusing on comedic techniques. First, comedic techniques were extracted, classified into categories, and a definition table was created. Subsequently, data on comedic techniques was collected using the comedic technique categories. Applying chi-squared tests and residual analysis to the collected data, the results suggested that there was a bias in the appearance trends of comedic techniques over time and that they were closely related to the development of the narrative.

　Magical girl genre has undergone diverse transformations from the 1960s to the present day and have established themselves as a representative genre within Japanese animation. Previous research on magical girl genre has primarily focused on humanities-based analyses, such as gender theory and sociological perspectives, and attempts to quantify narrative structures and conduct quantitative analyses have been insufficient. This study aims to quantitatively elucidate the evolution of narrative structures and world settings in 23 magical girl anime series produced and broadcasted in Japan from the 1960s to the 2020s. As an analytical method, we first performed hierarchical clustering on the world settings of the works, classifying the target series into three major clusters. Furthermore, N-gram analysis was applied to the narrative function categories within each cluster to extract their distinctive characteristics. The results of the analysis indicate that magical girl works have evolved from a structure where magic is used as a means of solving everyday problems to one centered on combat and conflict involving multiple characters.

　The most important and indispensable task in organizing regional archives is the creation of metadata. Until now, completing metadata has been an essential prerequisite for publishing regional archives. This research shifts that paradigm, proposing that metadata can be finalized after an archive's publication. Based on this concept, we developed a system that creates simplified metadata using AI prior to publication, then refines it after the archive goes live. We named this new method the “Myokenji Model”, after the region where it was first tested.

　Public administrations are increasingly developing web accessibility policies and publishing annual web accessibility test results for their websites. At the 2023 Annual Conference of the Japan Society for Information and Knowledge, a research presentation was given on the current state of web accessibility policy development and test result publication in Japanese cities, titled "Web Accessibility Policy Development and Annual Test Results Publication in Japanese Cities." Subsequently, research presentations were also given on the current state in towns and villages. Several years have passed since those research findings were published, and updates are expected, particularly regarding the publication of annual test results, which are scheduled to be conducted every year. This study focuses on towns and villages in Japan and aims to clarify the current status of web accessibility policy formulation and publication of annual test results related to web accessibility as of 2026.

　This paper reports on the evaluation and integrated design of two AI-assisted systems for building multilayered support systems for people in need: AskHow, which accumulates tacit knowledge as a graph, and the Mutual Aid Matching Assistant, which proposes collaboration plans. An evaluation workshop held in January 2026 achieved top-2-box scores (6-7 on a 7-point scale) of 53.8% for AskHow and 62.1% for the Matching Assistant, and the knowledge graph grew by 66%. Action nodes were the most frequent, indicating participants prioritized practical know-how. Based on these .ndings, we designed an integrated ontology that places concrete Actions at the core, structures organizational visions, resources, and needs to clarify matching rationales, and incorporates collaboration design patterns for ideation support.

　In recent years in Japan, the working-age population has been declining in many regions due to a falling birthrate, an aging population, and outmigration, making it increasingly difficult to address regional challenges. In response, the Ministry of Education, Culture, Sports, Science and Technology (MEXT) has, in part, placed expectations on regional universities to contribute to solving these issues. In fact, many universities are strengthening research rooted in their local communities in order to demonstrate their own significance. On the other hand, from a bibliometric perspective, there has been little development of indicators that capture such region-oriented research activities. This is likely due to several challenges. First, it is difficult to accurately extract place names (toponyms) from texts and other materials. Second, it is also challenging to uniquely identify and disambiguate those place names. Ultimately, there will be a need for mechanisms to uniquely link place names and geographic information to the relevant papers and datasets. While this study takes such solutions into consideration, it first focuses on the accuracy of place name extraction and conducts an experimental study using a compact test dataset.

　Recently, the public release of digital archives of local materials has progressed rapidly, and numerous “disaster archives” documenting specific disasters have been established. While records focused on disasters are important, it is essential to archive information spanning a “continuous timeline”. including not only the time of and immediately after a disaster but also the periods before and after.in order to preserve local history and provide reference information for future recovery efforts.
　Tentative name: The “Noto Vault” concept proposed in this study aims to build a foundation for the collection, accumulation, and publication of regional archival information pertaining to the Noto Peninsula. In particular, the Noto Peninsula Earthquake that occurred on January 1, 2024, and the Noto Peninsula Heavy Rainfall in September of the same year caused significant changes to the terrain and townscapes within a short period. In this report, we will examine the structure and role of regional archives in disaster-affected areas, focusing on the “Noto Vault” concept.

　This paper proposes a character setting support method utilizing a "character profile" format to enhance productivity in narrative creation. While previous studies have addressed the support or automated generation of partial character attributes such as appearance, personality, and backstory, there has been insufficient research into synthetical support that simultaneously covers external settings, psychological characteristics, and social/interpersonal relationships. To address this, we developed a method to assist creators in constructing a comprehensive character image by providing potential setting items and illustrative examples. To evaluate the effectiveness of the proposed method, we developed a prototype system and conducted user studies with six experts, including professional manga artists, art university faculty, and screenwriters. This paper reports on the validation of the necessity of character profiling, the limitations of the evaluation system, and specific improvements identified through the analysis of actual creative workflows.

　The generation of academic resource data from scholarly materials . such as historical documents and folk artifacts . encompasses multiple processes, including metadata creation, image capture, data organization, and verification for publication, all of which require diverse specialized expertise. However, a shortage of academic professionals due to population decline and growing societal demand for rapid data publication have made it increasingly urgent to streamline and optimize these workflows. The authors have extensive experience in surveying regional and scholarly materials, generating data, and publishing digital archives. In response to the rapid advancement of generative AI technologies, they have been working to revise and optimize academic resource data generation workflows through the application of these new tools. This report presents an overview and key challenges of a newly proposed human-machine collaborative workflow, in which humans and AI systems work in concert to improve efficiency throughout the data generation and publication process.

　This study investigates the usage patterns of academic information search tools and the influence of experience throughout the research cycle (pre-, during, and post-research) in today’s digital environment. A survey of 25 LIS researchers indicated that “communication with researchers” and “web search” serve as fundamental infrastructures, maintaining high usage rates across all phases. In contrast, the usage of specialized tools like Google Scholar decreased significantly as research progressed, confirming a transition toward purpose-specific optimization of search behavior. Furthermore, the findings reveal that experiences, such as success in earlier phases or distrust in generative AI, function as “strategic filtering” in subsequent tool selection. This suggests that researchers autonomously change and refine their search strategies based on empirical feedback, adapting to increasing academic rigor.

　This paper presents a collection of 640 use cases of DIAS (Data Integration and Analysis System) datasets, collected between October 2021 and January 2026 by manually identifying data references based on multiple search strategies, primarily using Google Scholar. The collection is published as Mahalo Cards on the Mahalo Button system. Analysis revealed that data DOI citations appear primarily in reference lists, while non-DOI mentions frequently appear in acknowledgments. The self-citation rate was 16.7%, a relatively low figure, indicating broad uptake of DIAS datasets beyond their creators. Non-article use cases, including government documents and web articles, were also identified, which are not captured by academic literature databases.

　Classification work in libraries is an intellectual activity involving expert judgment. The same item may receive different classification numbers across libraries depending on institutional policies, user populations, the edition of the Nippon Decimal Classification (NDC), and MARC data sources. Using data from 60 libraries participating in the National Diet Library Union Catalog Network (Uni-Ca-Net), this study analyzes inter-library variation in NDC numbers at the three-digit level for 1,190 Japanese-language items. Five classification patterns are identified: Type I (perfect agreement), Type II (near-perfect agreement with a single outlier library), Type III (dominant majority), Type IV (two-way split), and Type V (dispersed). Type I accounts for 732 items (65.0%) and Type II for 37 items (3.3%), together comprising 68.3%. In Type II, approximately half of the outlier classifications correspond to the same main class (first digit of NDC) as the majority, while the remainder fall into different classes. In 20.1% of cases, NDL assigned classification numbers different from those of other libraries; in Type II, 7 cases (18.9%) involved the NDL as the sole outlier. These findings guide recommendations for designing gold standard criteria for evaluating automated and AI-based classification systems.

　The Yi script, one of China’s ethnic minority writing systems, has often been noted for its graphic similarity to Chinese characters, and it has been argued that it may have been influenced by them. However, such claims have relied on subjective visual judgments, and quantitative verification has remained a challenge. To address this issue, this study employs machine learning to quantify glyph similarity by calculating the cosine similarity between embedding vectors derived from features learned from character images. In addition, because previous studies compared only a small number of arbitrarily selected characters, a systematic comparison was conducted using a vocabulary list designed for writing-system research. The results show that, aside from numeral characters, only a few Yi characters exhibit high similarity to Chinese characters. This suggests that it is difficult to posit a systematic borrowing relationship between the two writing systems, and that borrowing from Chinese was likely limited to numeral characters. Furthermore, these characters show particularly high similarity to Chinese regular script, suggesting that they may have been borrowed during the period when that script style was widely used.

　This paper positions the 100-year technological transition of IBM within the 50,000-year history of globalization and the 5,000-year genealogy of scientific civilization. It examines how information theory has defined social governance. In response to the inequality highlighted by J. Stiglitz’s "Economics of Asymmetric Information," the IT governance implemented by IBM offers a solution through data transparency. Furthermore, in the era of "probabilistic superposition" brought about by quantum computing, this paper redefines "NEXT Politics" as the exercise of "Observational Sovereignty," where humans determine meaning. Based on the spirit of "ONE FOR ALL, ALL FOR ONE," it argues that the "Active Choice" to delegate half of the processes to AI and quantum systems is the path to achieving the 22nd-century "Quantum Sophiocracy."

　This study analyzes the knowledge structure of history education content, focusing on the high school subject “Advanced World History,” specifically Major Topic B(2) “Characteristics of Ancient Civilizations.” The study draws on the curriculum guidelines and the descriptions in multiple history textbooks to extract key concepts such as “Civilizations,” “Agriculture,” “Environment,” “Culture,” “City,” “Social Status,” “Royalty,” “Religion,” and “Economy”. The relationships among these concepts－including compositional, causal, and chronological connections－are organized and visualized as a concept network, allowing us to examine how the educational content is systematically structured. Particular attention is paid to the centrality of the concept of “civilization” and its role in connecting other concepts, in order to investigate how learning about ancient history can foster the cognitive, evaluative, and expressive skills necessary for social participation. Through this approach, the study demonstrates the methodological potential of knowledge-structure analysis for research in history education.

　To clarify the impact of the mandatory high school informatics curriculum on the data science literacy of first-year university students, this study analyzed the trends in correct response rates from 2023 to 2025 at a small-scale regional university consisting solely of humanities and social science departments. Analysis revealed a dramatic improvement in correct answer rates for “Classification of Qualitative and Quantitative Data,” demonstrating the clear effectiveness of the new National Curriculum Standards in knowledge retention. Conversely, the correct answer rate for “Data Integration” remained low, indicating a persistent gap between the acquisition of isolated knowledge and the ability to synthesize multiple pieces of information. Based on the identification of these “skill gaps,” this report examines the future direction of data science education design at small-scale liberal arts universities.

　This study compares international trends in climate change education with Japan's current situation to examine the importance of ESD-based school management. As climate-related disasters intensify, UNESCO’s "Whole-School Approach" calls for integrated action across governance and curriculum. However, Japan’s meteorological disaster preparedness remains less structured than its earthquake measures. Utilizing the "Greening curriculum guidance" and the "Comprehensive School Safety Framework 2022-2030," this paper emphasizes integrating climate resilience into all educational activities. Moving forward, I intend to analyze practical crisis management, such as school closure criteria, to ensure alignment with these international frameworks.

　This study proposes an automatic generation method for "character backstory-inclusive stories." It focuses on "character backstories" (stories of a character's past) that create reader empathy and narrative inevitability, considering the reception mechanism of these backstories. In this paper, we generated plots based on frequent 3-grams of story structures extracted from seven analyzed works. Furthermore, we constructed an appearance probability model for character backstories, enabling the generation of backstory chains and backstory-inclusive plots. Specifically, this integration was achieved by assigning backstory categories to each plot scene based on the co-occurrence likelihood between story structures and backstory categories. An evaluation experiment involving 15 participants, including experts, revealed that stories generated by the proposed method received statistically significantly higher ratings across all evaluation items－such as "Are character backstories effectively narrated?"－compared to standard stories generated by existing algorithms.

　In manga, conspicuous frames known as catch frame are sometimes placed at important points in the story to draw the reader's attention. However, the techniques used to create catch frames and the reasons for using them are unclear. In this study, we focused on "Black Jack" and "Demon Slayer," and constructed a dataset to identify catch frames, classify the reasons for their appearance, and the manga techniques used. We also annotated the emotions of expected reader, protagonists, and other characters. Residual analysis of the chi-square test revealed that, despite differences in genre, there are commonalities between the two works.