Complex problems do not just ask for better answers; they ask for better ways of thinking. Accordingly, complex socio-technical design problems require integrated approaches that simultaneously address technical contradictions and human-centered processes. This study introduces contradiction-oriented exploration (COREX), a dual-track methodology designed to solve complex design problems involving both technical systems and human behavior. This approach combines two powerful tools: (i) The General Theory of Powerful Thinking-Theory of Inventive Problem Solving, which focuses on identifying and resolving system-level contradictions; and (ii) The Six-Box Scheme, which provides a user-centered, process-based framework for creative problem solving. By linking contradiction analysis with recursive exploration and real-world testing, this approach helps teams move from unclear user needs to structured innovations. The method was applied in a research and development setting focused on adaptive seat design. Participants followed a procedure that included problem modeling, contradiction identification, and inventive solution development. Results showed that COREX helped teams address design trade-offs more effectively than when using either method alone. The feedback cycles allowed for continuous improvement and system refinement. Overall, the methodology offers practical value for design teams working in emerging socio-technical domains by supporting both analytical thinking and creative ideation in an integrated process.

As a large language model, ChatGPT’s ability to learn from big data and respond to diverse user queries makes it a powerful tool for research and development. Despite the potential benefits of using ChatGPT, there are risks concerning users’ data protection. To address this issue, this study proposes utilizing Function-Oriented Search (FOS), a methodology based on Theory of Inventive Problem Solving (TRIZ). FOS provides an innovative approach to problem-solving by functionally defining a problem and generating solutions from areas where the function can be optimally performed. Thus, this study argues that applying FOS when using ChatGPT can ensure accurate results while mitigating the exposure of sensitive information. Although implementing FOS requires specialized training and sufficient hands-on experience to identify and conceptualize problem focus areas, ChatGPT can serve as an efficient tool for developers adopting this methodology. For both experts and novices in FOS, ChatGPT enables users to conduct efficient and comprehensive problem explorations and devise solutions. By demonstrating the application of FOS in practical cases, the study’s findings support the potential benefits of ChatGPT as a dynamic collaborator in problem-solving. The findings also indicate that FOS can guide the use of ChatGPT to generate suitable solutions while maintaining the protection of personal or corporate information. Overall, this study contributes to the emerging field of artificial intelligence by illustrating the possible synergy between TRIZ-based FOS and ChatGPT, a large language model.

Automatic text summarization (ATS) has gained increasing significance in recent years due to the rapid growth of textual data across digital platforms. The main objective of ATS is to generate a concise, informative summary from a lengthy document. Multi-document and multilingual summarization has been largely underexplored in previous research. This study presents an improved ensemble learning-based ATS system with slang filtering, using the Hyperfan-IN multilayer extreme learning machine-based autoencoder (HIN-MELM-AE) and the improved Dehghani poor-and-rich optimization algorithm (DePori). The original text undergoes comprehensive preprocessing, after which slang is detected and removed using DePori. Subsequently, the clean text is processed through info-squared C-means clustering, latent Dirichlet allocation-based topic modeling, term frequency-inverse document frequency weighting, and frequent-term extraction. Next, part-of-speech (POS) tagging is performed using a sememe similarity-induced hidden Markov model, and key entities are extracted from the transformed and POS-tagged data. Distilled bidirectional encoder representations from transformers (DBERT) are used to convert these entities into vectors. The final summary is generated through a combination of HIN-MELM-AE, stack autoencoder, variational autoencoder, and DBERT models, followed by cosine similarity calculation, voting-based fusion, re-ranking, and selection of the optimal sentences. Experimental results indicate that the proposed framework achieves superior performance 97.92% of the time, outperforming existing ATS methods.

Artificial intelligence (AI) and gesture recognition offer new creative possibilities, yet culturally sensitive, real-time systems for gestural folk music composition remain largely undeveloped. This study develops an AI-collaborative folk music composition system that integrates computer vision-based gesture recognition with specialized folk music generation algorithms to create a real-time interactive framework that supports traditional music composition while preserving cultural musical characteristics across multiple folk traditions. The system employs a four-layer architecture encompassing gesture acquisition, computer vision processing, interpretation, and generation layers. A comprehensive dataset of 1,643 folk music compositions from established repositories representing English, American, Irish, and Chinese traditional music (Nottingham Dataset, Irish Traditional Corpus, and self-recorded materials) was curated, supplemented by 6,127 successfully tracked gesture samples collected from 47 participants across 12 folk music gesture categories. The evaluation framework assessed gesture recognition accuracy, cultural authenticity preservation, real-time performance, and collaborative effectiveness through extensive experimental validation. The system achieved robust gesture recognition performance with 88.9% accuracy and 23.4 ms processing latency, while maintaining end-to-end response times of 86.8-91.6 ms during collaborative sessions. Cultural authenticity scores ranged from 7.6 to 8.3 across different regional folk styles, with a user satisfaction rating of 7.8 and a 28% improvement in musical coherence compared to baseline approaches. The framework successfully supports up to eight concurrent users while maintaining sub-100 ms real-time performance requirements. The integrated system successfully demonstrates effective coordination between gesture recognition and folk music generation subsystems, validating the architectural design and optimization strategies for culturally sensitive AI applications across diverse folk music traditions. The validated framework provides a foundation for educational, performance, and cultural preservation applications, contributing methodological insights for multimodal human-AI interaction systems and culturally aware creative technologies applicable to traditional music contexts.

As engineering systems accumulate increasing layers of functional, structural, and behavioral complexity, the ability to guide their evolution with coherent, theory-driven frameworks has become essential. This paper presents a cyclical theory of inventive problem solving (TRIZ)-based roadmap for the evolution of brushless direct current (BLDC) motors, guiding development from short-term corrective actions to long-term transformative strategies. The approach structures action into three coupled cycles that respectively prioritize rapid technical remedies, system-level contradiction resolution, and strategic system transition, enabling engineers to align interventions with the maturity and scope of each design challenge. It fuses core TRIZ instruments with the trends of engineering system evolution to couple contradiction handling with forward trajectories of system ideality. Applied to automotive BLDC applications, the method organizes recurrent issues such as acoustic anomalies, modal coupling, thermal stress, and control-layout interactions into an actionable roadmap that scales from quick design adjustments to modular, artificial intelligence-enabled capabilities. Experimental validation confirms the method’s practical impact: acoustic noise in the H24 configuration decreased by approximately 13%, modal vibration in the H8 case reduced by nearly 28%, and rotational imbalance amplitude in the rotor-yoke assembly dropped by around 55% after structural and dynamic optimization. The resulting framework is both prescriptive and extensible, guiding short-term fixes without foreclosing mid-term harmonization or long-term transformation, and generalizes to electromechanical product families that must balance cost, noise, durability, and intelligence under evolving requirements.