Stock market volatility research has long been the focus of industry and academia, and stock trend forecasting is challenging. Existing research focuses more on how to aggregate historical price features into graph networks, ignoring the effects of other information such as news and current events on forecasts. Most existing graph-based learning methods create stock graphs by manually constructing stock relationships, ignoring the complexity of stock relationships. Based on these, we propose a novel multilevel graph attention network (ML-GAT) for predicting the stock market trends. To be more specific, we initialize the node representations captured by each feature extraction module, then update the stock nodes in the isomorphic graph converted from Wikidata using ML-GAT, selectively aggregate the information of various relation types, aggregate the information to the node representation, and finally feed the result to a particular forecast layer for forecasting, completing the trend forecast of 423 stocks in the S&P 500 index and 286 stocks in the CSI 300 index. By comparing 5 popular approaches, the experimental research verifies ML-GAT’S state-of-the-art performance in prediction tasks. In comparison with the best performance benchmark model, the F1-score and accuracy increased by 11.82% and 12.6%, respectively, and the average daily return and Sharpe rate rose by 5.06 percent and 94.81 percent, respectively. More significantly, our model’s stock linkages are interpretable and consistent with real-world interactions.

Automatic detection of textile defects is a crucial factor in improving textile quality. Fast and accurate detection of these defects is key to achieving automation in the textile industry. However, the detection of textile defects faces challenges such as small defect targets, low contrast between defects and the background, and significant variations in the aspect ratio of defects. To address these issues, this study proposes a new method for textile defect detection based on an improved version of YOLOv8(You Only Look Once Version 8) called DA-YOLOv8s. Deep & Cross Network(DCNv2) is introduced into the Backbone Network to replace the C2F module, enhancing the extraction of network features; an self-attention mechanism, Polarized Self-Attention(PSA), is adopted to increase feature fusion capability and reduce feature loss in both channel and spatial dimensions; finally, a Small Object Detection Head (SOHead) is added to improve the feature extraction ability for small targets. Experimental results show that the improved YOLOv8 algorithm achieves has achieved mAP@0.5 and mAP of 44.6% and 48.6% respectively, which is an improvement of 4.2% and 3.8% over the original algorithm, and also outperforms the Optimal YOLOv9s model and the latest YOLOv11s model in these two metrics. The speed of textile defect detection has reached 257.38 frames per second (FPS) and the floating-point operation speed is 36.6 GFLOPS, ensuring the accuracy and speed of textile defect detection, with practical engineering application value.

Falling is an emergency situation that can result in serious injury or even death, especially in the absence of immediate assistance. Therefore, developing a model that can accurately and promptly detect falls is crucial for enhancing quality of life and safety. In the field of object detection, while YOLOv8 has recently made notable strides in detection accuracy and speed, it still faces challenges in detecting falls due to variations in lighting, occlusions, and complex human postures. To address these issues, this study proposes the SDES-YOLO model, an improvement based on YOLOv8. By incorporating a multi-scale feature extraction pyramid (SDFP), occlusion-aware attention mechanism (SEAM), an edge and spatial information fusion module (ES3), and a WIoU-Shape loss function, the SDES-YOLO model significantly enhances fall detection performance in complex scenarios. With only 2.9M parameters and 7.2 GFLOPs of computation, SDES-YOLO achieves an mAP@0.5 of 85.1%, representing a 3.41% improvement over YOLOv8n, while reducing parameter count and computation by 1.33% and 11.11%, respectively. These results indicate that SDES-YOLO successfully combines efficiency and precision in fall detection. Through these innovations, SDES-YOLO not only improves detection accuracy but also optimizes computational efficiency, making it effective even in resource-constrained environments.

As dynamic systems, social networks exhibit continuous topological changes over time, and are typically modeled as temporal networks. In order to understand their dynamic characteristics, it is essential to investigate temporal community detection (TCD), which poses significant challenges compared to static network analysis. These challenges arise from the need to simultaneously detect community structures and track their evolutionary behaviors. To address these issues, we propose TCDA-NE, a novel TCD algorithm that combines evolutionary clustering with convex non-negative matrix factorization (Convex-NMF). Our method innovatively integrates community structure into network embedding, preserving both microscopic details and community-level information in node representations while effectively capturing the evolutionary dynamics of networks. A distinctive feature of TCDA-NE is its utilization of a common-neighbor similarity matrix, which significantly enhances the algorithm’s ability to identify meaningful community structures in temporal networks. By establishing coherent relationships between node representations and community structures, we optimize both the Convex-NMF-based representation learning model and the evolutionary clustering-based TCD model within a unified framework. We derive the updating rules and provide rigorous theoretical proofs for the algorithm’s validity and convergence. Extensive experiments on synthetic and real-world social networks, including email and phone call networks, demonstrate the superior performance of our model in community detection and tracking temporal network evolution. Notably, TCDA-NE achieves a maximum improvement of up to 0.1 in the normalized mutual information (NMI) index compared to state-of-the-art methods, highlighting its effectiveness in temporal community detection.