Home Archive
Archive
2026 Volume 63 Issue 10  Published: 2026-10-01
    Special Issue on Security Protection Technology of Massive Heterogeneous Terminals in Dynamic Heterogeneous Networks
  • Rongxi Wang , Guanxiong Ha , Chunfu Jia , Keyan Chen
    doi: 10.7544/issn1000-1239.202660423

    Dynamic heterogeneous networks have become the mainstream paradigm for next-generation networks. The explosive integration of massive heterogeneous smart terminals imposes stringent requirements on encrypted data storage and privacy preservation within the cloud-edge collaborative architecture. Encrypted deduplication technology can identify and eliminate duplicate encrypted data, effectively reducing storage costs for cloud-edge environments while preserving data privacy, thus holding high practical value. However, existing encrypted deduplication schemes suffer from limitations such as vulnerability to duplicate faking attacks, high computational overhead, and heavy storage burdens on edge nodes, which fail to meet practical application demands. To address these issues, we propose SEDedup, a cross-user encrypted data deduplication scheme designed for cloud-edge storage scenarios. First, a server-aided encryption method suitable for cloud-edge storage is designed to achieve efficient and secure cross-user encrypted deduplication while effectively defending against duplicate faking attacks. Second, leveraging the temporal locality of data from heterogeneous terminals, an epoch-based duplicate data detection method is designed to achieve high-performance duplicate detection while minimizing the storage overhead on edge nodes. Finally, by incorporating generalized deduplication technology, lossless encrypted deduplication for similar heterogeneous terminal data is realized to maximize storage efficiency. Security analysis and performance evaluation demonstrate that SEDedup outperforms existing schemes in terms of performance while effectively preserving data privacy.

  • Special Issue on Security Protection Technology of Massive Heterogeneous Terminals in Dynamic Heterogeneous Networks
  • Hongzhan Ma , Ruofei Qu , Muyan Shen , Menghan Jia , Dengguo Feng , Yu Qin , Wei Feng
    doi: 10.7544/issn1000-1239.202660436

    The memory encryption mechanism in confidential computing environments renders the internal state of confidential virtual machines (CVMs) invisible to the host. While ensuring internal data security, this poses severe technical challenges for traditional hypervisor-based disaster recovery and backup solutions. Meanwhile, existing solutions relying on TEE secure processors suffer from severe performance bottlenecks, which heavily restricts the deployment of cloud confidential computing clusters and degrades their security resilience. We present ResiCVM (resilient confidential virtual machine (VM)), a resilient disaster recovery and backup framework based on intra-VM privilege isolation. ResiCVM shifts the state extraction and cryptographic protection logic from the legacy TEE secure processor to a high-privilege domain within the VM, strictly excluding the guest operating system from the trusted computing base (TCB). Additionally, we design an end-to-end cryptographic authentication protocol to ensure secure communication between the VM and the security management center during the state backup process. We implement a prototype of ResiCVM on the AMD EPYC 9745 platform. Experimental evaluations show that ResiCVM achieves a backup encapsulation and storage throughput 62.7 times higher than that of TEE secure processor-based schemes. Crucially, during concurrent multi-VM maintenance tasks, ResiCVM maintains stable execution efficiency, whereas TEE-based solutions suffer proportional performance degradation. Requiring zero host-side modifications, ResiCVM significantly improves disaster recovery performance and provides robust technical support for the secure operations and security resilience of confidential computing clusters.

  • Special Issue on Security Protection Technology of Massive Heterogeneous Terminals in Dynamic Heterogeneous Networks
  • Dajiang Chen , Yitao An , Lihua Jiao , Yuxuan Xiao , Fuhu Deng
    doi: 10.7544/issn1000-1239.202660259

    With the widespread deployment of IoT terminal devices in scenarios such as industrial control, intelligent transportation, smart homes, and smart healthcare, massive amounts of semi-structured logs are continuously generated at the edge. Existing methods either rely on log template IDs and shallow sequence models, making it difficult to model complex semantics, or rely on large-scale pre-trained models, which cannot meet the constraints of real-time performance and resource consumption of terminal devices. To address these issues, we propose a log anomaly detection technology for IoT terminal devices. First, a semantic enhancement detection model based on BERT and Llama is designed, employing regularization replacement and windowing modeling to suppress dynamic parameter noise. A semantic projection module is used to align the representation spaces of the encoder and decoder, and a three-stage fine-tuning method is used to improve the sequence-level anomaly detection capability. Second, a cloud-edge collaborative detection framework for terminal devices is constructed, proposing a two-stage confidence distillation and latency-sensitive dynamic routing strategy. This allows the small edge model to undertake most of the low-cost detection tasks, with only low-confidence samples uploaded to the large cloud model for verification. Our experiments on the public datasets HDFS, BGL, and Thunderbird-mini show that the proposed semantic detection model achieves F1 scores of 99.7%, 91.6%, and 96.6%, respectively. The cloud-edge collaborative framework maintains an overall F1 score of 95.5% while requiring only an average cloud routing ratio of 26.6%, which is equivalent to reducing the amount of large model calls by 73.4%.

  • Special Issue on Security Protection Technology of Massive Heterogeneous Terminals in Dynamic Heterogeneous Networks
  • Wei Yu , Guangquan Xu , Rujing Liu , Jinyao Zhang , Chunzhi Li , Hongpeng Bai
    doi: 10.7544/issn1000-1239.202660455

    To address resource constraints of terminal devices, distributed cross-domain trust, and dynamic access policies in large-scale Internet of things (IoT) environments, we propose a lightweight access control mechanism based on blockchain and fog computing. The mechanism adopts a layered architecture of on-chain trusted state maintenance, fog-side real-time authorization, and terminal-side lightweight identity response. High-complexity operations in attribute-based encryption (ABE), including key generation, policy matching, encryption, and decryption, are offloaded to fog nodes, while terminals only perform physical unclonable function (PUF) response generation and SHA-256 hashing. To support fine-grained and trusted cross-domain access, a hybrid authorization framework combining ciphertext-policy ABE (CP-ABE) and key-policy ABE (KP-ABE) is designed, where CP-ABE expresses resource-level access policies and KP-ABE supplements user-level resource permissions. Device identity hashes, policy states, and revocation events are recorded on a Hyperledger Fabric consortium blockchain, forming a trusted identity chain that combines PUF-based physical uniqueness with on-chain state consistency. In addition, a short-lived cross-domain access token and a challenge-hash-based verification method are introduced for cross-domain authentication and lightweight outsourced decryption verification. Security analysis shows that the mechanism satisfies identity authentication, session establishment, data confidentiality, collusion resistance, replay resistance, and unforgeability of cross-domain credentials. Prototype experiments show that, with five attributes, CP-ABE encryption and decryption take approximately 13.3 ms and 5.5 ms, respectively, and the data access path latency is approximately 18 ms. The results demonstrate that the proposed mechanism can support fine-grained, auditable, and cross-domain access control for resource-constrained IoT scenarios.

  • Special Issue on Security Protection Technology of Massive Heterogeneous Terminals in Dynamic Heterogeneous Networks
  • Chen Wang , Liqin He , Haoyu Ma , Jian Shen , Chunfu Jia
    doi: 10.7544/issn1000-1239.202660403

    State channels provide an effective solution for blockchain scalability by migrating transaction execution off-chain. However, existing approaches typically rely on strong assumptions, such as a static set of participants, continuously online nodes, and extremely low failure rates. These assumptions are difficult to satisfy in intelligent terminal environments and significantly hinder practical deployment. To address these challenges, we propose a state channel framework with offline proxy and fault-tolerant recovery for intelligent terminals. The main contributions are summarized as follows: 1) To mitigate channel reconstruction overhead caused by frequent joining and leaving of intelligent terminals, a dynamic membership management mechanism is proposed. Based on joint group signatures and symmetric credential verification, it enables members to complete permission changes without exposing real identities, achieving privacy-preserving and secure admission in dynamic environments. 2) To address transaction interruption caused by intermittent connectivity of intelligent terminals, a trusted execution environment (TEE)-assisted offline proxy mechanism is designed. Participants delegate transaction signing and state updates to a locally deployed TEE array, where offline proxy execution is performed in an isolated environment, ensuring continuity of blockchain operations in asynchronous and dynamic scenarios. 3) To cope with node failures or state loss risks, a fault-tolerant state recovery mechanism based on low-density parity-check (LDPC) codes is proposed. Participant states are encoded into sparse fragments and distributed across the TEE array, and sparse parity-check decoding is exploited to achieve high-probability and efficient fault-tolerant recovery. Experimental results demonstrate that, compared with existing methods, the proposed framework improves transaction success rates by 1.36–1.51 times, reduces latency by 20%–30%, and maintains system availability above 95% even with 20% state data loss, exhibiting significant performance advantages in dynamic intelligent terminal environments and partial failure scenarios.

  • Special Issue on Security Protection Technology of Massive Heterogeneous Terminals in Dynamic Heterogeneous Networks
  • Hui Zhou , Hao Chen , Zheng Qin , Yu Liu
    doi: 10.7544/issn1000-1239.202660415

    Federated learning (FL) enables collaborative model training without sharing raw data, making it well suited for collaborative intelligence among massive heterogeneous terminals in dynamic heterogeneous networks. However, its open participation mechanism makes FL vulnerable to poisoning attacks, where malicious clients can degrade the global model by performing label-flipping, backdoor injection, or model update manipulation. Existing poisoning defense methods primarily rely on robust aggregation strategies to mitigate malicious updates. Nevertheless, under data heterogeneity, these methods often suffer from limited adaptability, inaccurate anomaly detection caused by single-metric evaluation, and the loss of useful information due to hard filtering strategies. To address these issues, a unified poisoning defense framework for federated learning under data heterogeneity is proposed. The framework adopts ConTrust, a multidimensional consistency scoring-based defense method, as the foundation, which identifies malicious updates by jointly evaluating update direction consistency, update magnitude deviation, and principal subspace residuals. For non-independent and identically distributed (Non-IID) scenarios, HistTrust, a history behavior modeling-based defense method, is developed by incorporating local neighborhood references and cross-round reputation information, thereby mitigating the misclassification caused by data heterogeneity and improving malicious client identification. Experiments are conducted on MNIST, Fashion-MNIST, and CIFAR-10 datasets under multiple poisoning attack settings. Experimental results demonstrate that the proposed defense framework consistently reduces the attack success rate while maintaining high model accuracy under different data distributions, outperforming several representative defense methods in overall performance. The study indicates that designing differentiated defense mechanisms for different data distributions can effectively enhance the robustness of FL against poisoning attacks.

  • Special Issue on Security Protection Technology of Massive Heterogeneous Terminals in Dynamic Heterogeneous Networks
  • Zhiyuan Jiang , Gang Yang , Yongjun Wang , Zheng Qin
    doi: 10.7544/issn1000-1239.202660484

    In dynamic heterogeneous networks, massive heterogeneous terminals result in dispersed and diverse security data. While federated learning enables collaborative intrusion detection across terminals without compromising data privacy, its open architecture is inherently vulnerable to Byzantine attacks, where malicious clients poison the global model by uploading crafted gradients. Existing Byzantine-robust methods are typically effective only against specific attack types and struggle to cope with the coexistence of multiple attacks in real-world environments. Moreover, they suffer from high false-negative rates under sparse real-traffic conditions, and the computational overhead of stacking multiple defenses is prohibitive for resource-constrained terminals. To address these challenges, we propose GF-FedIDS, a Byzantine-robust federated learning intrusion detection framework based on gradient fingerprint decision-making. By analyzing the gradient statistics of typical Byzantine attacks, we establish a deterministic mapping from attack patterns to optimal defenses and reveal their causal relationship. A gradient fingerprint decision tree is designed to adaptively select a single optimal defense strategy, achieving full-coverage robust aggregation at unit computational cost. Furthermore, a class-aware instance weighting mechanism is introduced to mitigate missed detections under data imbalance. Extensive experiments on three public datasets and eight attack types demonstrate that, in single-attack scenarios, the attack detection rate (ADR) of GF-FedIDS outperforms that of the best baseline across three datasets; in mixed-attack scenarios, ADR reaches 1.45–2.14 times; and under extreme imbalance with a 5% attack rate, ADR is 2.06–2.29 times that of the best baseline. Additionally, GF-FedIDS incurs only a 16% computational overhead compared with FedAvg, outperforming existing ensemble defense methods.

  • Special Issue on Security Protection Technology of Massive Heterogeneous Terminals in Dynamic Heterogeneous Networks
  • Zinuo Cai , Haiwei Zhu , Qirui Wang , Hongrui Yin , Ruhui Ma
    doi: 10.7544/issn1000-1239.202660474

    In dynamic heterogeneous networks, massive heterogeneous terminals face a severe communication bottleneck when performing distributed collaborative learning: the frequent exchange of large-scale model parameters among terminals incurs heavy communication overhead. To address this issue, we propose QStar, a quantum-native federated aggregation algorithm for heterogeneous terminals. Unlike existing quantum federated learning schemes that convert quantum model updates into classical bit strings before transmission, QStar performs the transmission and aggregation of model updates within the quantum domain: clients encode local parameter updates into quantum superposition states so that a single quantum state transmission carries diverse candidate update strategies, and the server performs projective measurements guided by historical optimization trajectories to complete aggregation directly within the quantum domain. This mechanism makes the cost of carrying multiple candidate updates grow logarithmically rather than linearly with the number of candidates, and embeds candidate comparison and selection into the quantum measurement itself. Extensive experiments in typical heterogeneous terminal collaboration scenarios show that, compared with quantum federated learning baselines based on classical parameter transmission, QStar improves convergence stability and attains higher classification accuracy than parameter-quantization-based schemes on tasks with more classes and more complex features, providing a feasible technical pathway for collaborative learning of massive heterogeneous terminals in dynamic heterogeneous networks.

  • Special Issue on Security Protection Technology of Massive Heterogeneous Terminals in Dynamic Heterogeneous Networks
  • Yiming Wu , Jiamei Chi , Chenghao Chen , Zhenyu Wen , Zhen Hong , Xinwei Yao , Shouling Ji
    doi: 10.7544/issn1000-1239.202660432

    External vulnerability knowledge is increasingly important for large language models (LLMs) to understand vulnerability targets, craft exploit payloads, and interpret verification evidence. Recent approaches incorporate external vulnerability knowledge via prompt, task planning and multi-agent collaboration, to improve automated penetration testing. However, automated vulnerability reproduction remains challenging because public exploit materials are often heterogeneous and loosely structured, spanning scripts, HTTP requests, command-line snippets, and walkthrough in natural language. Prior work has leveraged retrieval-augmented generation, PoC semantic repair, and exploit code generation to better utilize external vulnerability knowledge, but still constrained by environment dependencies, exploit semantic extraction and model hallucinations, weakening the trustworthiness of automated reproduction.

    To address these issues, we propose ValidPoC, an LLM-driven framework for automated vulnerability reproduction that can convert heterogeneous exploit walkthrough or scripts into structured and reliable PoC-Nuclei templates. Specifically, ValidPoC defines a PoC semantic model that provides a unified representation of exploit logic, including request construction, payload generation, execution dependencies, and verification criteria. We develop a Semgrep-assisted semantic parsing mechanism to extract PoC Semantic IR from public exploit materials, including exploit scripts and natural-language descriptions, and construct structured Nuclei YAML templates for automated vulnerability reproduction. To improve reproduction reliability, we establish a closed-loop execution–evaluation–repair framework and a three-level validation model (L1–L3) for cross-validation across multiple execution evidence.We evaluate ValidPoC on 67 real-world vulnerability reproduction tasks. ValidPoC achieves a Pass@3 reproduction rate of 76.12% and consistently surpasses existing automated penetration testing frameworks, including PentestAgent and PentestGPT. In addition, it maintains competitive efficiency with respect to both execution time and token usage. These results demonstrate the effectiveness of ValidPoC in understanding heterogeneous exploit materials, generating PoC templates, and vulnerability reproduction.

  • Special Issue on Security Protection Technology of Massive Heterogeneous Terminals in Dynamic Heterogeneous Networks
  • Yao Liu , Lanyi Zhang , Mengyao Lu , Rui Hou , Ruilin Jiang , Yubo Wen , Yi Ding
    doi: 10.7544/issn1000-1239.202660465

    With the deep integration of cloud-edge-end collaborative computing and IoT, the massive heterogeneous terminals in dynamic heterogeneous networks have become the core protection objects of cybersecurity systems. The multivariate time-series data generated by these intelligent terminals during continuous operation, including industrial control sensor logs and network traffic features, carry complete semantic information about terminal behavior. Accurate classification of such time-series data is a critical foundation for terminal threat detection. However, the data distribution imbalance caused by terminal heterogeneity poses three key challenges to existing time-series classification models. First, constrained by fixed-resolution processing paradigms, models struggle to capture multi-granularity threaten features across different time scales. Second, existing methods forcibly map all resolution features into a unified high-dimensional space, ignoring intrinsic differences in feature dimension distribution across time resolutions, introducing noise, feature redundancy, and limiting deployment feasibility on resource-constrained edge terminals. Third, performing joint classification with multiple resolutions at inference incurs significant time and memory overhead from the feature extraction and fusion of each resolution branch, further exacerbating the inherent tension between lightweight deployment and high-accuracy detection. To address the alignment between time resolutions and feature dimensions, we propose an adaptive multi-resolution representation learning (AdaMRL) model for dynamic heterogeneous networks, achieving bidirectional alignment and decoupling of time resolutions and feature dimensions. The core contributions of this work are as follows. First, a multi-resolution Matryoshka feature architecture is constructed, equipping each time resolution with an independent Matryoshka representation learning layer to adaptively select an appropriate dimensional subspace for representation learning. Second, a multi-resolution joint loss supervision mechanism is proposed, which applies synchronized constraints to representations at each resolution during training to enhance cross-resolution feature robustness and accelerate model convergence. Third, an adaptive dimension selection and ensemble strategy is designed, dynamically matching the optimal feature dimensionality to each resolution at inference, with collaborative decision-making via soft-voting. Extensive experiments on 10 UEA multivariate time-series benchmark datasets and the industrial IoT edge security dataset Edge-IIoTset demonstrate that AdaMRL achieves state-of-the-art classification performance. Compared with baseline models, the proposed method improves average classification accuracy by 0.78% and reduces redundant feature parameter computation by 27.3% under optimal adaptive configuration, striking an effective balance between detection accuracy and computational efficiency, and providing robust support for real-time security protection of edge heterogeneous terminals.

  • Artificial Intelligence
  • Zhihong Wang , Wei Huang , Dexian Wang , Tianrui Li
    doi: 10.7544/issn1000-1239.202550613

    Anomaly detection for time series holds practical value in diverse fields such as financial risk control and industrial fault diagnosis, where the core challenge lies in constructing a representation space that effectively distinguishes between normal and anomalous patterns. However, the inherent characteristics of temporal data, coupled with the extreme scarcity of labeled anomalies, make it highly difficult to learn a representative mapping function with strong discriminative power, often leading to limited generalization capabilities. Meanwhile, contrastive representation learning constructs a discriminative feature space that clearly differentiates between instances, offering a more natural and promising solution for temporal anomaly detection. To address these challenges, we propose a dual anomaly-aware Transformer (DAT) model. By incorporating anomaly injection and anomaly awareness mechanisms, DAT is specifically enabled to capture the complex features of potential anomalous samples during the training phase. Additionally, a contextual contrastive loss function is adopted to guide a dual-branch structure for extracting more intrinsic and discriminative semantic features between normal and anomalous samples. Comprehensive experiments conducted on five public datasets demonstrate the superiority of DAT over 21 baseline models. The results show that DAT not only achieves competitive performance across all benchmark datasets but also achieves state-of-the-art (SOTA) performance on the MSL and PSM datasets.

  • Artificial Intelligence
  • Yaoping Hu , Shujuan Peng , Xin Liu , Zhen Cui , Lei Zhu
    doi: 10.7544/issn1000-1239.202550857

    Cross-domain graph anomaly detection aims to identify anomalies in a target-domain graph with limited labels, by leveraging both normal and anomalous samples from a source-domain graph. However, significant discrepancies in graph structures and feature distributions between different domains hinder effective knowledge transfer and limit cross-domain generalization. Moreover, the scarcity of labeled data in the target domain often induces ambiguous decision boundaries and unstable predictions during training. To address these challenges, we propose a cross-domain graph anomaly detection method that integrates dynamic prototypes with self-supervised transfer learning. Specifically, semantic prototypes are first constructed from the normal and anomaly samples in the source domain, and a bidirectional guidance strategy is then employed to align these prototypes with target-domain nodes in a shared semantic space, thereby alleviating cross-domain distribution shifts. To further enhance the stability and semantic discriminability, an exponential moving average mechanism is introduced for dynamic prototype updating. Subsequently, entropy minimization is applied to target-domain node embeddings to improve feature discriminability and reduce decision boundary ambiguity. Finally, pseudo anomaly scores can be well obtained by combining source-domain model predictions with unsupervised clustering results on the target domain data. Accordingly, high-confidence samples are selected as week supervision signals to iteratively refine the model, thereby improving both detection accuracy and robustness. Experimental results demonstrate that our method consistently outperforms state-of-the-art approaches in various cross-domain graph anomaly detection tasks, while validating its effectiveness and stability in cross-domain transfer scenarios.

  • Artificial Intelligence
  • Zhiliang Hu , Si Chen , Dahan Wang , Yan Yan
    doi: 10.7544/issn1000-1239.202550737

    Recently, the one-to-one label assignment rule has played a great role in removing non-maximum suppression (NMS) in the post-processing step and building an NMS-free end-to-end detection paradigm. However, this strict sample matching rule leads to a significant reduction in the number of positive samples during the training process, making it difficult for the model to fully mine the latent semantic information in the data during the feature representation learning stage, resulting in low learning efficiency. To this end, we propose a cross-label assignment knowledge distillation (CAKD) method for end-to-end object detection. This method combines the transfer learning mechanism of knowledge distillation with the end-to-end object detection algorithm to make up for the defects of the existing one-to-one label assignment rule and establishes an effective knowledge transfer path for the knowledge distillation framework to train end-to-end detection models. Specifically, the multi-scale features of the student model are first transferred to the detection head of the teacher model across models. Then, the student-teacher cross-label assignment predictions are compared with the teacher’s predictions to calculate the distillation loss. This method avoids directly forcing the student to imitate the teacher’s features, which leads to feature confusion and semantic misalignment. In addition, we also design an effective task-aware matching metric that comprehensively considers the quality of classification and regression, thus avoiding the problem of irrelevance between classification and localization tasks. We conduct a large number of experiments on COCO dataset to demonstrate the effectiveness of this method. By applying the cross-label assignment knowledge distillation method proposed in this paper to the non-end-to-end FCOS baseline model, it achieves a 38.8% non-NMS detection performance in terms of mean average precision (mAP), which is a 2.2 percentage point accuracy gain compared with the original detection performance with NMS.

  • Artificial Intelligence
  • Zefan Zeng , Yuehang Si , Qing Cheng , Zhong Liu
    doi: 10.7544/issn1000-1239.202550803

    Event causality identification (ECI) aims to detect causality between events in textual contexts. Compared with sentence-level ECI which focuses on single sentences, document-level ECI (DECI) needs to process more complex contexts such as long texts and cross-sentence events. Current DECI approaches face two key challenges: learning high-quality context-aware event representation to capture event-specific causal semantic and effectively integrating structural and semantic information for causal reasoning. To address these challenges, we propose MAIGO (multi-granularity context-aware and iterative graph optimization), a novel DECI framework that achieves context-aware event representation through multi-granularity attention mechanisms and dynamically optimizes the event causality graph structure via iterative embedding refinement. MAIGO enhances event embeddings through sentence-level, document-level, and graph-level attention, capturing event-event, event-sentence, and sentence-sentence interactions to produce better representation for events. The framework further constructs an event causality graph (ECG) and jointly optimizes both the ECG structure and event embeddings via masked attention in an iterative optimization process. Extensive experiments and analysis on two benchmark datasets and a multilingual dataset demonstrate MAIGO’s effectiveness, achieving performance gains of 4.1% (causality existence identification) and 7.3% (causality direction identification) in inter-sentence ECI over state-of-the-art models. Further analysis confirms the advantages of the multi-granularity context-aware enhanced representation strategy and iterative graph optimization mechanism.

  • Artificial Intelligence
  • Li Lin , Zongsheng Li , Shuai Wang
    doi: 10.7544/issn1000-1239.202550842

    Inventory scheduling plays a critical role in modern supply chains, with its primary objective being to balance production and demand through appropriate inventory allocation. As market demand becomes increasingly uncertain and product attributes more diverse, traditional experience-based inventory scheduling methods often struggle to effectively integrate product characteristics with demand fluctuations. Moreover, their decision outcomes lack interpretability, making it difficult to trace the rationale behind inventory schedules and reducing stakeholders’ trust. The semantic understanding and knowledge-reasoning capabilities of pre-trained large language models, when combined with abundant real-world multi-source external data and multi-category demand data, offer opportunities to enhance existing LLM-based inventory scheduling methods. To capitalize on these opportunities, LLM-based inventory scheduling decision tasks have to face the following challenges: 1) how to effectively embed the inventory scheduling workflow into a multi-step reasoning process, yielding accurate and coherent reasoning trajectories; 2) how to balance the interpretability of allocation decisions with decision accuracy, avoiding a decline in interpretability caused by excessive focus on accuracy. To this end, we propose a novel multi-step reasoning framework for inventory scheduling based on large language models, termed InvLLM, with following primary contributions: the design of a self-calibrating multi-step reasoning method based on the inventory scheduling workflow and a multi-objective balanced optimization method for inventory scheduling decisions. Additionally, to address distortion in the reasoning fine-tuning phase, a “planning-token” supervised fine-tuning approach is proposed, combined with a generative reward model, achieving a balance between high accuracy and logical consistency. Experimental evaluations on a large-scale real-world dataset provided by JD.com demonstrate that, compared with existing inventory scheduling methods, the proposed InvLLM framework increases service rate by 3.24% and improves overall inventory evaluation scores by 1.55%, while reducing inventory rate by 1.33%, thereby validating its effectiveness and practical applicability.

  • Artificial Intelligence
  • Bowen Dong , Xueli Liu , Xiaoyang Feng , Zhenzhen Mai , Wenjun Wang
    doi: 10.7544/issn1000-1239.202550845

    Knowledge graph reasoning has attracted extensive attention because of its ability to alleviate the inherent incompleteness of knowledge graphs. However, existing approaches mainly rely on embedding models or chain-based rules, but they still suffer from limitations in interpretability and expressiveness. Existing rule learning methods are often limited to linear dependency modeling and struggle to capture multi-path interactions and hierarchical semantic relations, while traditional rule-mining methods usually follow the “enumeration–verification” paradigm, resulting in high computational costs and insufficient modeling of deep semantic correlations among rules. To address these challenges, we propose CGRL (contrastive graph rule learning), a novel method for learning graph-structured rules through contrastive learning. CGRL enhances rule learning through a threefold mechanism. First, it applies an adaptive clustering strategy to generate soft type labels for entities, introducing semantic priors that improve rule generalization. Second, it designs a multi-hop neighborhood sampling method to construct semantic subgraphs and to generate high-quality positive and negative rule instances. Third, it employs a Transformer-based encoder and a contrastive learning objective to map rule pairs into a unified embedding space, thereby jointly capturing structural dependencies and semantic consistency. Extensive experiments on six public knowledge graph completion benchmarks demonstrate that CGRL achieves substantial improvements over state-of-the-art baselines. On datasets with relatively complex relational structures, such as FB15k-237 and WN18RR, CGRL yields average gains of 6.04% in MRR and 16.26% in Hits@1, confirming its effectiveness in discovering high-quality graph-structured rules and significantly enhancing reasoning performance.

  • Artificial Intelligence
  • Tingting Li , Shouqiang Sun , Qingqing Li , Ling Wan
    doi: 10.7544/issn1000-1239.202550930

    The recognition of negative sentiment intensity in financial public opinion on social media is of great significance for financial risk prevention and market stability. Given the prevalence of multimodal information combining text and images in financial sentiment expression, and the need for further exploration of structural correlations between modalities, we propose a multimodal negative sentiment intensity recognition model based on the collaborative enhancement of graph convolutional networks and Stacking ensemble learning. The model first employs the CLIP model to extract semantic features from text and images. Subsequently, a multimodal graph structure is constructed and graph convolutional networks are introduced to achieve structural association embedding of features, thereby enhancing the ability to capture potential inter-modal correlations. Finally, a two-layer Stacking ensemble learning framework is built to optimize the decision performance of sentiment intensity recognition by integrating the prediction results of multiple diverse base classifiers. Experimental results show that under the optimal configuration, the model achieves an F1-score of 83.57% and an AUC of 95.05%, outperforming the optimal single-text model, the optimal single-image model, and the original CLIP multimodal model by 2.48, 5.27, and 1.87 percentage points in F1-score, respectively. These results verify the effectiveness of the proposed method in mining modal correlations and improving recognition performance, providing technical support for intelligent financial public opinion analysis and financial risk monitoring.