The increasing prevalence of mental health disorders highlights the need for innovative and accessible interventions. Although existing digital meditation applications offer valuable basic guidance, they often lack interactivity, real-time personalized feedback, and dynamic simulation of real-life scenarios necessary for comprehensive experiential training applicable to daily stressors. To address these limitations, this study developed a novel immersive meditation system specifically designed for deployment within a metaverse environment. The system provides mindfulness practice through two distinct modules within the virtual world. The experience-based module delivers AI-driven social interactions within simulated everyday scenarios, with narrative content dynamically generated by large language models (LLMs), followed by guided inner reflection, thereby forming a scenario–experience–reflection cycle. The breathing-focused module provides real-time feedback through a breath-synchronization interface to enhance respiratory awareness. The feasibility and preliminary effects of this metaverse-based system were explored in a two-week, single-group, pre-test/post-test study involving 31 participants. The participants completed a battery of validated psychological questionnaires assessing psychological distress, mindfulness, acceptance, self-compassion, and self-esteem before and after engaging in the intervention. This study provides exploratory evidence supporting the feasibility and potential of immersive metaverse environments and LLM-based scenario generation for structured mental health interventions, providing initial insights into their psychological impact and user experience.
Emotion recognition in conversations is a key task in natural language processing that enhances the quality of human–computer interactions. Although existing deep learning and Transformer-based pretrained language models have shown remarkably enhanced performances, both approaches have inherent limitations. Deep learning models often fail to capture the global semantic context, whereas Transformer-based pretrained language models can overlook subtle, local emotional cues. To overcome these challenges, we developed EmoBERTa–CNN, a hybrid framework that combines EmoBERTa’s ability to capture global semantics with the capability of convolutional neural networks (CNNs) to extract local emotional features. Experiments on the SemEval-2019 Task 3 and Multimodal EmotionLines Dataset (MELD) demonstrated that the proposed EmoBERTa–CNN model achieved F1-scores of 96.0% and 79.45%, respectively, significantly outperforming existing methods and confirming its effectiveness for emotion recognition in conversations.
Procedural approaches have long been used in game development to reduce authoring costs and increase content diversity; however, traditional rule-based systems struggle to scale narrative complexity, whereas recent large language model (LLM)-based methods often produce outputs that are structurally invalid or incompatible with real-time game engines. This gap reflects a fundamental limitation in current practice: generative models lack systematic mechanisms for managing executable game knowledge rather than merely producing free-form narrative texts. To address this issue, we propose a Game Knowledge Management System (G-KMS) that reformulates LLM-based narrative generation as a structured knowledge management process. The proposed framework integrates knowledge grounding, schema-governed generation, normalization-based repair, engine-aligned knowledge admission, and application within a unified pipeline. The system was evaluated on a compact 2D Unity-based RPG benchmark using automated structural and semantic analyses, engine-level playability probes, and a controlled human player study. The experimental results demonstrated high reliability in knowledge admission, stable procedural structures, controlled expressive diversity, and a strong alignment between system-level metrics and player-perceived narrative quality, indicating that LLMs can function as dependable knowledge-construction components when embedded within a governed management pipeline. Beyond the evaluated RPG setting, this study suggests a practical and reproducible approach that may be extended to other executable systems, such as interactive simulations and training environments.
Game commentary enhances viewer immersion and understanding, particularly in football video games, where dynamic gameplay offers ideal conditions for automated commentary. The existing methods often rely on predefined templates and game state inputs combined with an LLM, such as GPT-3.5. However, they frequently suffer from repetitive phrasing and delayed responses. Recent studies have attempted to mitigate the response delays by employing traditional machine learning models, such as SVM and ANN, for event prediction. Nonetheless, these models fail to capture the temporal dependencies in gameplay sequences, thereby limiting their predictive performance. To address these limitations, an integrated framework is proposed, combining a lightweight convolutional model with multi-scale temporal filters (OS-CNN) for real-time event prediction and an open-source LLM (LLaMA 3.3) for dynamic commentary generation. Our method incorporates prompt engineering techniques by embedding predicted events into contextualized instruction templates, which enables the LLM to produce fluent and diverse commentary tailored to ongoing gameplay. Evaluated in the Google Research Football environment, the proposed method achieved an F1-score of 0.7470 in the balanced setting, closely matching the best-performing GRU model (0.7547) while outperforming SVM (0.5271) and Transformer (0.7344). In the more realistic Balanced–Imbalanced setting, it attained the highest F1-score of 0.8503, substantially exceeding SVM (0.4708), GRU (0.7376), and Transformer (0.5085). Additionally, it enhances the lexical diversity (Distinct-2: +32.1%) and reduces the phrase repetition by 42.3% (Self-BLEU), compared with template-based generation. These results demonstrate the effectiveness of our approach in generating context-aware, low-latency, and natural commentary suitable for real-time deployment in football video games.
Simulation-based testing is fundamental for validating automated driving systems (ADSs), and real-world traffic accident records provide realistic scenario data. However, converting such records into executable simulations requires addressing structural incompleteness, geographic bias in large language models (LLMs), and code-generation errors. This paper proposes an adaptive ontology–LLM framework that generates Scenic scenarios executable in CARLA from structured accident records obtained from the Korea Road Traffic Authority’s Traffic Accident Analysis System (TAAS). The framework comprises five modules connected through explicit interfaces and feedback loops: Ontology-Based Data Structuring, Missing Attribute Inference, Korean-Context Filtering, Scenic Scenario Code Generation, and Simulation Validator. Only Korean-Context Filtering is country-specific, supporting future adaptation by replacing this module. Across three complete executions of the final pipeline with the map-topology and feasible-placement pre-checks applied, Ontology-Based Data Structuring, Missing Attribute Inference, and Korean-Context Filtering achieved mean success rates of 91.00%, 92.37%, and 94.07%, respectively. The mean Step 3 cumulative success rate was 78.30%, and the mean end-to-end success rate was 61.83% (range: 59.00–63.50%). Korean-Context Filtering achieved an F1 score of 0.9246 on the Original Korean-Context Evaluation Set and a mean F1 score of 0.8708 on the Additional Korean-Context Evaluation Set. The results demonstrate the feasibility of integrating ontology mapping, causal reasoning-based attribute inference, geographic-bias mitigation, and executable scenario generation in a unified framework.