Computer Network Information Center

Corporate profiling serves as a critical analytical tool for modern enterprises, enabling data-driven decision-making in investment strategies, risk assessment, and strategic planning. It requires integrating quantitative metrics, qualitative insights, and network relationships to capture a company’s role in the business ecosystem. However, traditional methods struggle to synthesize heterogeneous data and model complex interdependencies among corporations, news, and market dynamics, often addressing these aspects in isolation. To address these challenges, this article introduces Financial Graph-based Mixture of Experts Prompt Learning (FGMPL), an innovative framework that unifies graph prompt learning with a multi-task paradigm for corporate profile modeling. The proposed framework reformulates node- and edge-level tasks into a coherent graph-level representation and employs multi-view contrastive learning to effectively integrate textual details with relational structures. Moreover, a novel Financial Multi-Experts Prompting mechanism—with learnable tokens coupled with a Mixture of Experts (MoE) design—is presented to enhance the processing of heterogeneous graph data and bridge the gap between pre-training and downstream tasks. To further improve adaptability, a meta-learning-based prompt tuning strategy is incorporated, enabling rapid transition to various downstream applications. Extensive experiments on real-world financial graphs show that FGMPL consistently outperforms strong pre-training and graph-prompting baselines across corporate performance prediction, relationship prediction, and news classification in both full-data and few-shot settings. In addition, cross-market transfer on a NASDAQ dataset and interpretability/efficiency analyses further demonstrate its robustness and practical applicability.

Abstract Oceanic meridional heat transport (MHT), a key component of Earth's energy flow, fundamentally shapes climate variability by redistributing heat between hemispheres and across latitudes, thereby regulating surface energy exchange, atmospheric circulation, and hydrological patterns. However, accurate quantification of the global MHT remains challenging. Here, we use the energy budget approach to derive MHT from 1985 to 2023 based on sea surface heat fluxes (Fs), ocean heat content tendency (OHCT), and heat changes related to sea ice melt/formation (Q ice ), with objective adjustment applied to these components under an ocean energy budget constraint. This approach enables an improved estimate of the MHT climatology, variability, and trend for the global, Indo-Pacific and Atlantic oceans. With good agreement with RAPID and OSNAP mooring array observation time series, our estimates indicate realistic northward MHT at all latitudes in the North Atlantic with a peak of 1.22 ± 0.07 PW at 24.5°N. In contrast, the Indo-Pacific MHT is poleward with a peak of −1.67 ± 0.06 PW at 13.5°S. Regional energy budgets reveal that Fs dominates the mean state of MHT, while OHCT controls its interannual variability. After 2000, when the data quality of both OHC and Fs improved a lot, the Indo-Pacific Ocean exhibited a statistically significant increase in MHT, whereas the Atlantic Ocean showed a basin-wide weakening of MHT, particularly between 25°S and 18°N. The derived ocean MHT data in this study provide a basis to evaluate model and reanalysis data, and support a better understanding of the Earth’s energy flow.

Abstract Accurately simulating orography‐induced mountain waves over steep terrain, such as the Tibetan Plateau (TP), remains a major challenge for numerical weather prediction (NWP) models due to grid distortions inherent in traditional terrain‐following coordinates. To address this issue, we developed an AI‐driven adaptive mesh refinement (AMR) framework within the Fluidity‐Atmosphere model, which employs a 3D unstructured mesh to mitigate geometric distortions. A Long Short‐Term Memory (LSTM) neural network is integrated to enhance the AMR process, replacing traditional adaptation criteria with data‐driven predictions. A series of idealized 2D and 3D experiments demonstrate that both the traditional AMR and LSTM‐driven approaches reproduce mountain wave dynamics with higher efficiency than fixed mesh. Furthermore, the LSTM model suppresses numerical noise near terrain, preventing spurious over‐refinement. In realistic simulations over the TP, the LSTM‐enhanced model successfully captured the life cycle of mountain waves, reproducing key physical features such as vertical velocity structures, wave amplitude decay, and upstream phase tilt. Comparative tests further revealed efficiency gains of up to 71.4% over fixed meshes and 23.8% over traditional AMR at high resolution, alongside accuracy improvements in vertical velocity, potential temperature, and wave propagation. These findings validate the LSTM‐based AMR framework as a robust and efficient approach for atmospheric simulations over complex terrain. By intelligently allocating computational resources while preserving physical accuracy, this method offers a scalable pathway toward next‐generation atmospheric modeling, with future applications targeted at realistic meteorological conditions over the TP.

Cognitive modeling, as an emerging technology in the field of computer-aided education, aims to explore students’ knowledge levels and learning abilities to achieve various intelligent educational applications. Although some existing work focuses on addressing the problem of student forgetting, it is still a less explored area how to naturally integrate the forgetting effect caused by the time interval between answering exercises into student knowledge state modeling. Additionally, traditional cognitive modeling methods mostly assume that students answer exercises one by one, which often does not align with real answering behavior and cannot be directly extended to diverse learning scenarios. Therefore, in this article, we propose a Continuous Time-based Neural Cognitive (CT-NC) framework and several implemented models (CT-NCM and two extensions) to effectively integrate the dynamic and continuous characteristics of knowledge forgetting into student learning process modeling, making it more natural. Specifically, we adopt a specially designed learning event encoding method to adjust the neural Hawkes process to capture the relationship between knowledge learning and forgetting over continuous time. Furthermore, we propose a customizable learning function to jointly model the changes in different knowledge states and their interaction with each practice moment. In the end, we demonstrate an extension CT-NCM+ that can adapt well to diverse learning scenarios, indicating that CT-NCM can solve real-world problems by flexibly adjusting its structure. Extensive experimental results on real datasets clearly demonstrate that CT-NCM and CT-NCM+ outperform the current state-of-the-art KT methods in student performance prediction, while our work points out a realistic research direction for KT and demonstrates its interpretability in knowledge learning visualization.

Small target detection in Unmanned Aerial Vehicle (UAV) applications is often plagued by inherent challenges such as small object sizes, sparse information, and complex background interference. Traditional detection algorithms and existing YOLO series models suffer from limitations in detection accuracy and fine-grained detail preservation. To address this, this paper proposes YOLO-DSNet, a small target detection network based on YOLOv13n. First, we introduce the dual-stream attention module (DSAM), which enhances discriminative features by leveraging bidirectional context modeling. Second, we design the Multi-scale Attention C2f (MSA-C2f) module—an adaptive architecture that optimizes feature extraction via multi-scale enhancement, effectively preserving and integrating small target information. Finally, through dataset augmentation, we significantly improve the model’s detection performance. The proposed YOLO-DSNet achieves a mAP@0.5 improvement from 30.8% to 40.1% on the VisDrone2019 dataset with only 0.8 million additional parameters, yielding a 30% accuracy gain while increasing computational overhead by merely 11.6 Gigaflops (GFLOPs). Experiments demonstrate YOLO-DSNet’s effectiveness in small target detection tasks such as UAV aerial photography and remote sensing imagery, successfully balancing accuracy and efficiency with high practical value.

<title>Abstract</title> To address the critical challenges in the accurate "expert-project" matching within intelligent review scenarios—specifically the dynamic drift of experts’ research interests over time and the inadequacy of single-modal information in characterizing complex scientific research content—and to solve the issue where existing sequential recommendation methods fail to adapt to sudden interest changes in real-time due to offline training, this paper proposes an Intelligent Review Sequential Recommendation Algorithm based on Reinforcement Learning and Multi-modal Interest Drift (MM-DRLSR). Closely aligning with the construction requirements of the intelligent review system for the National Independent Software and Hardware Ecosystem Verification Public Service Platform, the algorithm first models the latent distribution of experts’ research direction drift through drift degree quantification and a disentanglement mechanism, employing a drift-aware alignment module to capture collaborative research relationships among experts. Secondly, a lightweight attention fusion module is designed to integrate multi-modal features, including texts and charts from project proposals, thereby enhancing project representation. Finally, combined with dynamic reinforcement learning, the matching strategy is optimized online utilizing review feedback. Experiments on public datasets demonstrate that MM-DRLSR achieves a 16.7% to 23.0% improvement in Recall and NDCG metrics over strong multi-modal baselines and adapts rapidly in sudden drift simulations. These results validate the algorithm’s effectiveness in resolving expert interest drift and precise matching issues, providing efficient algorithmic support for expert recommendation in intelligent review systems.

ABTRACT The discovery of topological materials is severely hampered by fragmented research workflows that cause information loss, inconsistent reasoning, and frequent computational failures. To overcome these barriers, we present TopoMAS, an interactive multi‐agent framework that unifies the entire discovery pipeline through human–AI collaborative intelligence. TopoMAS seamlessly integrates natural language processing, knowledge retrieval from literature and databases, crystal structure generation, and automated first‐principles validation. At its core, is a multi‐level reasoning and coordination mechanism coupled with a self‐refining knowledge graph. This architecture enhances query understanding and ensures computational robustness by adaptively allocating tasks, monitoring execution, and recovering from failures. In collaboration with human experts, TopoMAS has accelerated the identification of candidate topological phases and successfully guided the discovery of new materials. Benchmark evaluations show that TopoMAS's coordinated intelligence enables smaller, more efficient models to rival or even surpass the performance of substantially larger counterparts at a fraction of the computational cost. Ultimately, TopoMAS offers not only a powerful accelerator for materials research but also a transferable blueprint for building next‐generation, AI‐augmented discovery platforms across scientific disciplines.

Vegetation carbon turnover time (τveg) dominates the uncertainty in terrestrial carbon cycle dynamics. Reports have shown that the τveg of mature or old-growth forests in North America and Europe has decreased because of faster carbon loss under global climate change. However, the temporal trend of τveg in widespread younger forests, which exhibit different growth patterns, remains inconclusive. Here, we consistently revealed a significant overall increase in τveg (0.025±0.002years per year) across China's forests that are characterized by a relatively young forest age structure, using multisource data-model assimilation and long-term network observations. In young forests that receive high nitrogen deposition, increasing levels of CO2 accelerate vegetation growth, causing it to grow faster than it dies or decomposes, leading to increases in the τveg. The effects of forest age on τveg dynamics and the high sensitivity of enhanced carbon sinks to τveg should be incorporated into future land surface models to ensure that the τveg dynamics and their impacts on terrestrial carbon cycling and climate mitigation are accurately assessed. Our results also provide valuable insights for forest management aimed at enhancing carbon retention and sequestration through optimizing age-related stand dynamics.