Credit card fraud detection has become increasingly challenging because modern financial transactions are highly heterogeneous, temporally evolving, structurally connected, and severely imbalanced. To address these challenges, this paper proposes CTH-RiskGNN, a causal temporal heterogeneous graph neural network designed to improve credit card fraud detection by jointly modeling transaction relations, temporal risk evolution, causal stability, and fraud risk diffusion. The ablation analysis further confirms that heterogeneous relation modeling, temporal encoding, causal stabilization, risk-diffusion memory, gated fusion, and imbalance-aware optimization all contribute to the final performance improvement. These findings indicate that integrating causal temporal graph learning with heterogeneous risk diffusion can effectively overcome the limitations of independent transaction classification and provide a more robust framework for detecting concealed, evolving, and relational financial fraud.
Demand forecasting in fast-moving consumer goods (FMCG) supply chains is challenging: demand is volatile, promotions are frequent, and product portfolios are diverse. Forecast errors propagate upstream and intensify the bullwhip effect, leading to excess inventory, stockouts, and higher operating costs. To address these issues, we propose ILX-DFIO, an integrated two-stage framework in which a Long Short-Term Memory (LSTM) network models temporal dynamics and XGBoost corrects residuals using exogenous signals such as promotions, prices, and category attributes. The design targets accurate short-term predictions while maintaining robust performance across multiple horizons. Experiments on three real-world FMCG datasets against seven competitive baselines show consistent gains: relative to the strongest baseline, ILX-DFIO reduces RMSE by 9–15% overall, improves RMSE by up to 9.1% during promotional periods, and lowers total inventory cost by 5.5%.
The prevalence of social networks encourages employees to express their authentic selves, resulting in more frequent emotional fluctuations. Despite the importance of emotions for creativity, existing research neglects their dynamic effects on creativity within social networked environments. This study introduces 'emotional drift' and examines its impact on creativity across four studies. The results indicate that moderate negative emotional drift fosters creativity more than positive drift. As intensity increases, positive emotions enhance creativity while negative emotions inhibit it. Network heterogeneity moderates these effects—amplifying weak negative and strong positive drift's benefits, while buffering strong negative drift's harm. Additionally, organizational citizenship behavior stability relates to creativity nonlinearly, following an inverted U-shaped pattern, mediating the emotional drift–creativity relationship. This research integrates emotional dynamics into creativity research, offering insights for managing creativity in social networking contexts.
Social media comments provide timely and publicly observable feedback for brand marketing, yet their sentiment expressions are often short, informal, context-dependent, and semantically complex. Conventional polarity classification is therefore insufficient for detecting hidden dissatisfaction, sarcasm, and aspect-specific brand problems. To address this limitation, this study proposes LACoS-BERT-CRF, a large language model-assisted framework for complex sentiment analysis in social media brand comments. The framework integrates contextual sentiment classification, complex sentiment modeling, and CRF-based aspect-opinion extraction to support both affective judgment and interpretable brand feedback diagnosis. Experiments were conducted on four public datasets: Weibo Senti 100K, ToSarcasm, DiaASQ, and Amazon Dog Food Reviews. On Weibo Senti 100K, Chinese RoBERTa-wwm-ext achieved the best overall sentiment classification performance, with an Accuracy of 0.9781 and a Macro-F1 of 0.9781.
As enterprises generate vast amounts of heterogeneous knowledge, existing knowledge management systems face challenges in providing contextually relevant, accurate, and domain-specific answers. To address these issues, the authors propose KQGen, a unified framework that integrates knowledge graph embeddings with generative question answering for enterprise knowledge bases. KQGen encodes knowledge graph entities and relations, performs query understanding, and generates answers grounded in structured knowledge, optimized by a reinforcement learning-based feedback mechanism. Experimental results on benchmarks such as SQuAD, TREC, and KBQA show that KQGen outperforms existing models in answer relevance and ranking metrics. However, challenges remain in improving performance for Exact Match (EM) and scalability. Future work will focus on refining multi-hop reasoning, real-time scalability, and integrating multi-modal data for enhanced system adaptability.
