In real-world data integration scenarios, traditional equi-joins and other join techniques have huge difficulties due to heterogenity and inconsistencies in attribute values. To address this challenge, we present AutoStarJoin, a technique for automated joins specifically designed for star-join scenarios. The core contribution of our approach is the automated detection of join attributes across arbitrary schemas, reducing or even eliminating the need for manual specification. Our approach analyses first the edit-based distance measures, transforming similar string values within join attributes to facilitate matching, and explores then token-based distance measures, refining the join process by identifying optimal attribute pairs. We evaluate the effectiveness of various distance metrics in terms of join quality and computational efficiency across diverse datasets. Our goal is to generalize across different join scenarios without requiring domain-specific parameter tuning. This level of automation makes our approach suitable for integration into AutoML pipelines where minimal human intervention is desired. However, the correct choice of parameters per data set is crucial. We intend to implement hyperparameter optimization in further research.
Analyzing visitor flows in natural or touristic areas is key for assessing users’ preferences and behaviour, and for providing relevant input for infrastructure development and visitor management measures. Despite of the growing tendency of electronic device usage, in-person surveying still remains an invaluable approach for getting in-depth information on how visitors use and move along trails, alongside with their motivations, preferences and general mindset. An annual, survey-based visitor monitoring project was launched in 2017 in the UNESCO Pilis Biosphere Reserve in Central Hungary with the above aims, where trip routes are questioned from trail users and recorded via counting sheets. These data are compiled into a unified database, which provides a basis for a visitor flow map and various analyses. This paper aims to present the current state of the modeling and processing framework we have been developing for representing and processing the trip routes recorded in textual form and translated into trail graph routes, alongside with the key challenges and findings we have so far. Beyond its particular local relevance, our approach has revealed significant general route information modeling aspects and considerations that are useful when dealing with the duality of text-based, human-described geographical route representations and their (possible or assumed) exact geo-localized linestring representations.
The growing complexity of deep learning models introduces challenges in scalability and adaptability. This paper explores how modular design, inspired by software engineering, can enhance deep learning systems. This paper also showed that modern deep learning techniques such as Mixture of Experts (MoE) and LoRA are advancing toward higher modularity. By promoting modular architectures, we emphasize the need to move beyond monolithic models toward more reusable, maintainable, and scalable AI systems, providing a potential research direction for future work.
Applications of data science to football are blooming such as to calculate prediction of matches by applying machine learning. In this paper, we propose an implementation method of geometric analysis system for football that converts an actual coordinate data set into the system to geometrically calculate indicators describing game dynamism. The purpose of this project is to design and implement a system that dynamically extract knowledge of football in visible form. As an example of the indicators, Shooting Degree Value, SDV, is defined and implemented, and an example study is illustrated with the prototype system and an actual data set of J.LEAGUE.
As general-purpose LLMs become widely accessible, tailoring their outputs to domain-specific tasks remains challenging. This paper investigates how custom, pattern-guided datasets and fine-tuning can specialize LLM behavior. We run a four-round open-vs-closed comparison: (1) evaluate commercially hosted models with an open-source dataset; (2) repeat with a synthetic, pattern-aligned variant; (3) replicate both rounds with open-source LLMs; and (4) demonstrate a real-world case by fine-tuning an open model to answer WordPress technical support queries using a finalized, pattern-constrained dataset. Across rounds, fine-tuned open-source models achieved performance comparable to closed-source baselines while offering greater controllability and deployment flexibility. We detail dataset design, the role of response patterns in producing consistent outputs, and practical trade-offs observed during fine-tuning and real-world use. The findings indicate that with well-crafted custom datasets, open-source LLMs can be reliably specialized for targeted applications without reliance on proprietary APIs.
One of the essential computations in environmental study is “semantic computing” in a context-dependent way to analyze the changes of various situations and phenomena in a natural environments(water, air, CO2, places of livings, sea level, coral area, etc.). Humankind, the dominant species on Earth, faces the most significant mission; we must endeavor on a global scale to perpetually restore and improve our natural environments. It is important to realize global environmental computing methodology for analyzing natural phenomena and influences of nature and livings in a context dependent way with a large amount of information resources in global environments.We have presented a “Semantic Computing System” [1,2,3] for analyzing and interpreting the meanings of concepts, objects and phenomena, applied to natural environment-analysis and changes occurring in the world. As a global environmental system based on the semantic computing system, we have realized “5-Dimensional World Map System(5D World Map System)” [4,7,10] for integrating, analyzing, retrieving and visualizing environmental situations in oceans and landsThe 5D World Map System is globally utilized as a Global Environmental Semantic Computing System, in SDG-14, United Nations-ESCAP (https://sdghelpdesk.unescap.org/toolboxes). We have also proposed the concept of “Semantic-SPA (Sensing, Processing and Analytical Actuation Functions)” for realizing a global environmental system, to apply it to 5-Dimensional World Map System [4,7,10]. This concept is significant to design environmental systems with Cyber-Physical space-integration to detect environmental phenomena as real data resources in a physical-space (real space), map them to cyber-space with semantic computing, and propagate the analytical computing results to the physical-space with visualization for expressing environmental situations, phenomena, causalities and influences. It is also significant to memorize the situations and compute environmental changes in various contexts, in order to discover what are happening in the nature of our planet. We have various (almost infinite) aspects and contexts in environmental changes in our planet, and it is essential to realize a new analyzer for computing differences in those situations for discovering actual aspects and contexts in nature.
The goal of this paper is to implement the Transparent Intensional Logic framework in the Haskell programming language, based on the procedural, i.e., constructive nature of this logical system. The main benefits of this implementation include type checking of constructions for free and the possibility of using predefined functions of Haskell, which allows focusing solely on the essence of this work, namely, the transformation of various constructions of Transparent Intensional Logic into Haskell terms.
We propose a novel module for computing all possible passages through an intersection by assigning indexes to entry and exit points extracted from textual narratives and computing relative directions between them. Our method relies on a contiguity condition that ensures every input record is connected and computes relative directions using predefined tables. The resulting passages serve as the basis for constructing reliable sketch maps with topological accuracy. This approach facilitates early error detection in spatial data and supports future extensions to cases where intersections are incomplete by retaining non-contiguous records until additional data become available.
This paper proposes a method to optimize human resource allocation for tasks involving human interaction. The method considers task success rates under temporal and spatial constraints, human experience and skills, and contextual factors. Human interaction is modeled in three types: autonomous, distributed, and collaborative. Additionally, individual skills, past performance, and biases are incorporated as variables to reflect real-world complexity. The approach enables formalize resource allocation processes traditionally managed through human communication. This paper presents the method’s formalization and compares five computational approaches. Experimental validation was conducted within the domain of systems engineering, utilizing a sample dataset comprising 100 tasks T and 1,000 individuals P. The proposed method demonstrated high reproducibility and effectiveness in task resolution.
This paper proposes a new conceptual framework for modeling human action. There are too many aspects for describing human behavior, however, the paper proposes only seven dimensions that describe human experience. The first four dimensions describe physical characteristics consisting of 3D position and time. The remaining three dimensions describe physiological, psychological, and Social characteristics respectively. This model can be applied to the full range of human experiences. The methods for measuring each parameter of the seven dimensions are discussed. The process is exemplified in the Data Sensorium setup.
It is challenging for a computer to comprehend dynamic characteristics and variables of a human user; however, conceptual modeling can serve as a language to describe these dynamic aspects. In this research, we aim to model the human user’s varying input and output modalities in human-computer interaction, making the human appear as part of the system architecture. This allows technologies to adapt to the user’s different input and output modalities. As a first step, we model a class diagram to depict a systemic view of interaction. Secondly, we formulated a state machine diagram to describe the behavior of the user and IT artifact in the context of accessible interaction. These models contribute to accessible interaction between humans and computers, where the computer either changes or adjusts its own input and output modalities so that the human can receive them and provide their own outputs according to their abilities.
The meaning of a conceptual model is derived from the linguistic semantics of the modeling language used, the intension of the modeling concepts employed, and an explicit specification of these concepts that is understood within a community of practice. The triptych paradigm of conceptual modeling reflects these three dimensions of meaning as represented in the triptych wings of ‘Language Wing’, ‘Model Wing’, and ‘Meaning Wing’. In this paper, we investigate how the specification, that is, the ‘grounding’ of the modeling concepts, can be achieved based on these triptych wings. Our considerations should help the development of information systems to avoid costly future system changes that could occur due to the lack of a comprehensive understanding of an application and its inherent semantics during modeling. When developing a system, specifying the concepts used for modeling as precisely as possible increases the likelihood that the resulting information system will meet the application requirements.
This paper presents a paradigm shift in natural language processing through the evolution of our Knowledge Matrix theory into a coordinate-based direct mapping system with flow-controlled slicing operations. Moving beyond traditional deep learning approaches based on attention mechanisms and contextual representations, we establish a foundational mathematical framework where tokens map to scalar identity values and texts unfold as temporal sequences in a five-dimensional Text Space defined by time (t), token identity (y), sentence identity (s), paragraph identity (p), and text unit identity (e). This representation enables both text analysis and generation through straightforward geometric operations controlled by hierarchical flow patterns, eliminating the need for neural networks, attention mechanisms, and complex training processes. Flow control operates through a hierarchical E→P→S→Y selection cascade where each dimensional choice constrains and guides subsequent selections, creating natural text generation through geometric constraint propagation. Through practical implementation, we demonstrate our method’s effectiveness in text generation while maintaining perfect reconstruction fidelity. Experimental results validate both the five-dimensional framework and large-scale conversational applications achieving ChatGPT-like functionality through coordinate operations alone. Our approach suggests a fundamental rethinking of natural language processing, emphasizing mathematical clarity and flow-controlled geometric operations over architectural complexity.
This study proposes a novel method for quantitatively and visually representing individual preferences for piano performance styles. Such preferences are often highly subjective and difficult to articulate, thus limiting effective learning and the development of personalized artistic expressions. Using features such as tempo, dynamics, and their first- and second-order derivatives, and leveraging the Approximate Inverse Model Explanation (AIME) framework from Explainable AI (XAI), the method analyzes preference labels assigned by learners to professional performances, enabling the visualization of integrated, preference-based performance styles independent of specific pieces. This approach allows learners to explicitly visualize and verbalize their preferences, incorporate them into their performances, and identify professional performers whose stylistic tendencies align with their preferences. The experimental results demonstrate that the proposed method effectively captures integrated stylistic tendencies across multiple pieces and reveals stylistic similarities among performers that may remain hidden when using conventional numerical metrics alone. By making the abstract concept of performance style preferences concrete, the proposed method provides a systematic, explainable, and personalized framework that lays the groundwork for future studies aimed at bridging the gap between subjective musical preferences and actual learning outcomes, with potential applications in individualized music education, artistic self-discovery, and enriching musical performance culture.
Business processes represent an essential part of every business as they create the structure within which a company operates. In order to capture, understand, and improve these processes, business process models are created using various modeling notations such as BPMN. One of the key challenges is validating that these models reflect the reality of the business in a correct and accurate manner. Only then it is possible to develop information systems based on these models that truly support the workflow of their users. Based on our previous research, this paper presents a novel approach that enables the validation of business process models through simulation in graph databases. As a starting point for our approach, we propose a conversion of models expressed in BPMN into OntoUML, a modeling notation that provides a higher level of abstraction as well as unique integration of the process and structural perspectives on the problem domain. Conversion to OntoUML also enables us to utilize our tool for automated generation of integrity constraints for Neo4j graph databases, ensuring consistency and adherence to the defined rules. We are then able to validate these models by simulating object instantiation, progressively creating and modifying graph representations of entities and relationships to detect inconsistencies. This simulation not only reveals potential flaws in the original business process models but also offers actionable insights for their refinement. By shifting the focus to the underlying data and its lifecycle, our approach provides a novel perspective on business process model validation, also bridging the gap between models and their practical implementation.
The main goal of this paper is to propose a formalism for type checking constructions of Transparent Intensional Logic capable of detecting the widest possible range of type-theoretical incoherence. To achieve this goal, however, a slight modification to the type system of Transparent Intensional Logic is necessary, specifically in the specification of higher-order types. This adjustment brings its type system closer to the type systems of standard programming languages within the functional paradigm, thereby increasing the practical implementation potential of Transparent Intensional Logic, which has so far been primarily utilized as a syntactic standard.
With the growth of internet usage, countless educational videos are now available online. However, it can be a significant challenge for learners to identify the videos they need, especially in their preferred language and within their available time. Additionally, not all videos are suitable for subject-specific learning due to variations in length and presentation components. According to Sweller’s Cognitive Load Theory, working memory during the learning process is highly limited. Learners must be selective about which information from sensory memory they choose to focus on. In our proposed Co-Learning model (a model of connective learning where all necessary knowledge is refined and interconnected to support effective learning within cognitive limitations), we leverage NLP approaches to enhance the learning experience. These approaches include video speech refinement, subtitle generation, dubbed video translation, summarization, classification, keyword extraction for word cloud indexing, and quiz generation, thereby creating a multilingual, learner-efficient environment. In our preliminary survey, the generated content was well-received and effectively utilized for class adjustments with an acceptance rate of 93%.
The Paris Agreement of 2015 has prompted countries to accelerate their efforts to become carbon neutrality efforts, which meant reducing CO2 emissions to virtually zero. Limiting global warming to less than 1.5°C by 2050 rely on technologies that remove CO2 from the atmosphere faster than humans release it. This implies that CO2 will be removed at a rate of 1-30 gigaton per year by 2050. Carbon Capture Storage / Sequestration (CCS) and Utilization (CCU) are concepts and technologies that collect emitted CO2 store it permanently underground, or recycle it as energy or chemicals for use in manufacturing and other economic activities. CCS and CCU have been discussed globally, but have not reached local and practical levels. Currently planned large-scale CCS requires significant government investment and new technological developments for capture, transport, and storage / sequestration, therefore implementation is expected to start in the second half of 2030 towards the 2050 goal. The need to start acting now where possible rather than waiting for the distant future, makes it important to implement CCS and CCU on a small scale and build towards future scale-up as an immediate solution. This study proposes a support method and system to help companies that emit large amounts of CO2 such as power plants, cement, petrochemicals, and steel industries, to decide how to treat their CO2 emissions in the context of decarbonization. In this study, a Simple, Measurable, Attainable, Relative, and Time-Bound (SMART) decision support method and Direct Air Capture Location and Cost Simulator (DLCS) system were developed to provide a solution to the Negative Emission 5W1H “What, Who, Which, When, Why, and How” from the perspective of a company that emits CO2. A prototype model with parameter settings was proposed based on knowledge gained from practical experience. The functionality of the SMART method and DLCS system was confirmed by applying sample data from the actual data of the ‘Tokyo Region’ as a Proof of Concept (PoC). In this PoC, characteristics of direct air capture which is a critical technology for negative emissions, were verified. The core of the SMART and DLCS model entails combinatorial optimization, distance calculation, cost estimation, and market projection including constraint solution.
Every discipline – and even more so a new one – needs a practical and expressive theory. With the science and practice of modelling, we are devoting ourselves to such a lore and teachings of MMM. Here MMM is an acronym that stands for the three components of a theory and practice of model-based science: M(odels)–(to )M(odel and use)–M(odelling). With the third component, we focus on the systematic, well-thought-out and well-founded way of working with models and model activities that can be understood as the Grand Theory of Models. MMM is combined with a corresponding scientific and practical discipline that we call modelology. Everything can potentially assimilate model-hood, but some things can only be models to a very limited extent, if at all. There are also non-models, pre-pre-models, pre-models, proto-models, or even un-models. Being a model can be characterised as a sophisticated judgement based on the capability and validity of models, the potency of models, the embedding in the background and the application, as well as the journey of a model over time for changing communities of practice. This reliability judgement leads to a conceptual model of model-hood itself.
As they have been for decades, databases are a key component in various types of business applications. Nowadays, modern databases also include features that support the construction of applications based on artificial intelligence. The starting point of this paper is the questions of what are the typical use cases in which a database and generative artificial intelligence work together, what database features are used in these applications, and what other technologies the applications rely on. The results contribute to one of our practice-oriented research projects, in which we build an environment that supports AI experiments for participating companies. As input, we use open online learning materials provided by three database vendors. The databases we selected for review are all non-relational, representing three different genres, and are the most popular representatives of their genre. For each training program of the vendors, we will examine the structure and scope of the program, the courses on generative AI included in the program, and the use cases and technologies related to generative AI presented in them. Finally, we will prepare a summary of the use cases and technologies found.
