Home Archive
Archive
2026 Volume 63 Issue 3  Published: 2026-03-01
    Viewpoint
  • Guojie Li
    doi: 10.7544/issn1000-1239.202660032

    This paper reconceptualizes AI safety from the lens of logical complexity, and establishes three levels of safety issues: R1 Level (decidable propositions, supporting formal pre-verification), R2 Level (semi-decidable propositions, supporting only post-hoc evidence discovery), and R3 Level (non-recursively enumerable, where even the identification of insecurity cannot be guaranteed). Distinguishing between R1 and R2 is critical: all engineering-tractable security issues reside within the R1 Level. Consequently, achieving security requires a dual-track effort focused on correctness verification and institutional safety nets. Regarding the high-profile field of AI security, while current risks have not yet escalated to the R3 Level, the governance trajectory must urgently shift from “pre-verification” to “runtime governance”, prioritizing external monitoring including gating, rollbacks, isolation, human-in-the-loop systems, and permission hierarchies. Furthermore, it necessitates a dual-sovereignty system combining embedded technology and external institutions, thereby maintaining human corrective sovereignty and civilizational security within a logically incomplete reality.

  • Special Issue on Data Storage and Processing for New Quality Productive Forces
  • Jiwu Shu , Yijie Wang
    doi: 10.7544/issn1000-1239.1239.qy20260601
  • Special Issue on Data Storage and Processing for New Quality Productive Forces
  • Xi Cao , Mingjie Li , Chaoshu Yang , Cheng Yang , Runyu Zhang
    doi: 10.7544/issn1000-1239.202550474

    Emerging zoned namespace solid state drive (ZNS SSD) addresses several critical issues inherent in traditional SSDs, including high write amplification, low storage density, and complex I/O paths. These advancements create new opportunities for the progress of storage technologies. The B+tree, as an efficient tree index structure, is widely used in various databases and file systems to support the efficient loading of large models, the construction of external knowledge bases, and the management of structured metadata, thereby significantly improving training efficiency and knowledge invocation performance. However, due to the hardware characteristics of ZNS SSDs being different from traditional block devices, directly deploying B+tree to ZNS SSDs not only results in higher write amplification rates but also causes cascading updates, seriously affecting the performance of the storage system. For the above problems, we propose DPZB+tree, a B+tree indexing structure optimized for ZNS SSDs and incorporating persistent memory (PM). First, DPZB+tree adopts a hybrid DRAM-PM-ZNS SSD storage architecture, thereby effectively separating hot data and cold data. Second, DPZB+tree introduces a lightweight hot/cold node identification strategy, which in turn improves I/O efficiency. Third, to address the limited capacity of PM, we present a dynamic node placement strategy that adaptively migrates data between PM and ZNS SSD. Finally, we combine hardware characteristics and spatiotemporal locality principles to design leaf node splitting and merging operations. DPZB+tree is implemented using a ZNS SSD simulator and Intel optane PM. Experimental results show that under various workloads, DPZB+tree outperforms LSM tree, SSDB+tree, DZB+tree, and Baseline in terms of performance and recovery time.

  • Special Issue on Data Storage and Processing for New Quality Productive Forces
  • Qiang Li , Tao Liu , Xueqing Zhang , Xianwei Meng , Minggang Sun , Bole Li , Xi Chen
    doi: 10.7544/issn1000-1239.202550376

    Global cache is a core technology for enterprise-grade storage arrays, whose primary function is to improve data write performance. I/O operations return immediately once data is written into the memory cache of storage controllers, without waiting for the data to be persisted to slow HDDs or SSDs. However, the design of global cache in enterprise-grade storage products must comprehensively consider factors such as the number of replicas, system high availability, and failure scenarios. Constrained by the mutual interactions of these factors, the number of controllers in storage clusters typically cannot exceed 4, and system high availability fails to support the scenario where three out of four controllers fail concurrently. Furthermore, the transmission of cache replica data between controllers via RDMA(remote direct memory access) technology also impairs storage performance. Thus, a memory-disaggregated global cache system prototype is designed based on the latest CXL(compute express link) technology. With memory cache disaggregated from within controllers to external, the system offers three key advantages: first, controllers and memory cache can scale out independently, supporting multiple cache replicas; second, controller failures do not cause cache data loss, supporting concurrent failures of n−1 out of n controllers; third, CXL technology significantly optimizes storage performance while enhancing system scalability and high availability. Experimental results demonstrate that the CXL-based global cache achieves a 38% performance improvement compared with the RDMA-based global cache.

  • Special Issue on Data Storage and Processing for New Quality Productive Forces
  • Jian Gao , Jiwu Shu
    doi: 10.7544/issn1000-1239.202550361

    Distributed lock is a crucial component in distributed storage systems. The performance of the lock protocol significantly influences the overall performance of the entire system. Remote direct memory access (RDMA) is an emerging data center networking technology that supports one-sided communication verbs and offers low CPU overhead, low latency, and high throughput. It presents new opportunities for designing high-performance distributed lock protocols. However, designing such RDMA based network protocols faces significant challenges, for example scalability and fairness. We address these challenges by proposing FeLock, an RDMA-based distributed lock protocol. FeLock achieves high performance while tackling the above challenges by leveraging different RDMA communication verbs, which enables the clients to communicate directly with both the server for lock acquisition and release, and other clients to hand over lock ownership. Specifically, first, to improve performance, FeLock introduces a per-node lock management mechanism to reduce network roundtrips on the critical paths of the lock protocol. Second, to achieve scalability, FeLock incorporates a round-robin handover mechanism, in which nodes are logically organized into a ring, and clients hand over lock ownership sequentially according to their positions within the ring. Third, to ensure fairness and prevent clients from starvation, FeLock employs a node credit mechanism that limits the number of consecutive lock acquisitions by any single node, thereby preventing clients on any node from being indefinitely blocked by others. Experimental results demonstrate that FeLock achieves performance comparable to or exceeding that of existing one-sided RDMA lock protocols, such as DSLR, while exhibiting better fairness and scalability. With 3 to 120 clients, FeLock achieves throughput 1.01 to 7.51 times of DSLR, with its fairness improved by up to 2.24 times.

  • Special Issue on Data Storage and Processing for New Quality Productive Forces
  • Fangyu Wei , Tianyun Chen , Yang Zhou , Yulai Xie , Fang Wang
    doi: 10.7544/issn1000-1239.202550480

    In the context of the information age, the explosive growth of data volume and the diversification of compression application scenarios raise higher demands for the flexibility and efficiency of compression strategies. The LZ77 algorithm, a classical lossless compression method, is widely used in mainstream compression tools such as Zstandard (ZSTD). However, achieving higher compression ratios often requires larger history windows and more complex compression strategies, which in turn lead to frequent cache misses and reduced efficiency of lazy matching in ZSTD’s LZ77 implementation. To address these issues, two optimization strategies are proposed. First, the multi-level region search strategy (MLRS) introduces a hierarchical matching region mechanism along with access threshold control, enabling adaptive adjustment of search depth and restricting memory access range during the matching process to alleviate cache pressure. Second, the extended search-based lazy matching strategy (ESLM) reuses search paths and incorporates approximate substitution techniques to reduce redundant computation while improving matching efficiency. These strategies are implemented and evaluated under ZSTD level 12 configuration on Kunpeng 920 server platform. Experimental results demonstrate that MLRS significantly reduces last-level cache miss rates across various datasets and improves compression throughput to 118.34%−149.50% of the baseline while maintaining a compression ratio within 94.65%−99.58%. ESLM improves compression throughput to 113.49%−117.46% of the original while further increasing the compression ratio on most datasets. When both strategies are applied jointly, the compression speed increases to 134.53%−171.17% of the original scheme, while sustaining a compression ratio in the range of 94.18%−99.80%.

  • Special Issue on Data Storage and Processing for New Quality Productive Forces
  • Lun Ge , Yijie Wang , Ying Zhang , Xiaohui Zhou
    doi: 10.7544/issn1000-1239.202550364

    The rapid development of emerging industries such as artificial intelligence, big data, and cloud computing has led to the widespread application of large-scale cloud computing centers, which continuously generate massive amounts of operational data during operation. Efficient and accurate anomaly detection is a key part in ensuring intelligent operation and maintenance as well as the system reliability. Due to factors such as equipment aging, sudden environmental changes, and dynamic loads, operational data often contain problems such as data noise pollution, abnormal signal coupling, and dynamic drift in equipment status. Current mainstream anomaly detection methods typically rely on assumptions of data purity, identifying anomalies by learning normal data features and evaluating reconstruction errors. However, industrial-grade intelligent systems face inherent challenges such as sensor noise contamination, dynamic state drift, and coupled anomaly signals due to their complex topological structures and heterogeneous operating environments, all of which degrade data quality and compromise detection efficacy. To address these issues, we propose a time-series anomaly detection method based on time-frequency enhanced contrastive learning(TF-ECL). It introduces a dual-view enhancement mechanism to implement controllable augmentations in temporal and frequency domains, thereby improving model noise immunity based on cross-view contrastive learning. Meanwhile, based on the differential representations of abnormal signals in time-frequency space, a dual-domain collaborative detection mechanism is developed to achieve abnormal localization using time-frequency contrastive errors. Experimental results from multiple real-time data sets demonstrate that compared with previous multidimensional time-series anomaly detection methods, TF-ECL achieves robust and excellent performance, with metrics such as F1 score, VUS-PR, and VUS-ROC surpassing or matching those of current state-of-the-art (SOTA) methods.

  • Special Issue on Data Storage and Processing for New Quality Productive Forces
  • Zhihang Tang , Hao Lan , Zhengguo Liu , Qing Wang , Laiping Zhao , Keqiu Li
    doi: 10.7544/issn1000-1239.202550478

    With the continuous growth of the parameter scale of large language models (LLMs), fine-tuning models with hundreds of billions of parameters pose extremely high requirements for computing and storage resources. Traditional distributed training schemes often rely on a large number of high-end graphics processing units (GPUs) and high-speed interconnect networks, resulting in extremely high training costs. Although the existing single-GPU training schemes relieve the pressure on GPU memory through tensor offloading, they still face problems such as low I/O transmission efficiency and insufficient device utilization. Traditional kernel-space I/O operations introduce frequent system calls and context switching during large-scale tensor transfers, becoming a key bottleneck that restricts performance. Meanwhile, the optimizer computations fail to fully leverage the parallel capabilities of multi-core CPUs, making it difficult to effectively overlap with GPU computations, further limiting system performance. In response to the above problems, we propose HiTrain, a heterogeneous memory offloading and I/O optimization scheme for LLM training, with a focus on the design and implementation of two key technologies: First, it constructs a high-performance tensor storage module based on the storage performance development kit (SPDK), which manages tensor data in user space, thereby avoiding the overhead of the kernel I/O stack and improving the concurrency and throughput of tensor offloading. Second, it designs and implements an asynchronous-optimizer based storage-compute pipeline scheduling module, which reorders optimizer execution to reduce GPU idle time and thus enhance overall training efficiency. The experimental results show that on the server equipped with a single GPU and a non-volatile memory express solid state drive (NVMe SSD), the scheme proposed can make full use of the computing and storage resources in the system, improve the efficiency of tensor unloading and loading in the process of model training by 32.7%, and increase the overall training throughput to 1.49 times of the existing scheme, offering a practical and cost-effective path for LLM training.

  • Special Issue on Data Storage and Processing for New Quality Productive Forces
  • Liu Wang , Shenao Wang , Xinyi Hou , Jian Zhao , Rongxin Wu , Qiao Xiang , Yanjie Zhao , Yi Wang
    doi: 10.7544/issn1000-1239.202550481

    Large language models (LLMs), as a cornerstone technology in natural language processing (NLP), have demonstrated exceptional capabilities in text generation, information retrieval, and conversational systems. These models, such as ChatGPT, LLaMA, and Gemini, have been applied across various fields, including healthcare, education and finance, achieving near-human or even superhuman performance. However, with the widespread adoption of LLMs, their storage mechanisms face significant security and privacy risks throughout their lifecycle. Core storage modules, including model file storage, inference caching, and knowledge vector storage, support the functionality and efficiency of LLMs but also expose vulnerabilities and threats. For example, model file storage faces risks such as weight leakage and backdoor injection, inference caching is susceptible to side-channel attacks, and knowledge vector storage faces data poisoning and workflow hijacking. To address these emerging challenges, researchers have proposed defense strategies such as model encryption, backdoor detection, cache partitioning, and content filtering techniques. Although existing techniques have improved security to some extent, LLM storage technologies still face a range of critical challenges. This paper systematically reviews security risks and defense mechanisms for LLM storage across its lifecycle, focusing on attack surfaces and mitigation strategies. It identifies current limitations and highlights future research directions to enhance the security and reliability of LLM storage mechanisms.

  • Special Issue on Data Storage and Processing for New Quality Productive Forces
  • Jinjiang Wang , Xiao Zhang , Linxuan Su , Jinyang Yu , Zhijie Huang , Yiwei Gan , Helang He , Ngulube Michael
    doi: 10.7544/issn1000-1239.202550482

    Modern blockchain systems face increasing storage pressure due to the continuous growth of on-chain data volume. Most systems adopt LSM-tree-based key-value storage engines as their persistence layer, which maintains intra-level key order through compaction. However, blockchain data are highly heterogeneous, characterized by structurally diverse and unordered keys. This results in frequent compaction operations that cause write amplification and degrade I/O efficiency. To address these issues, we propose AppendChain, a storage optimization framework that reorganizes the write path and key-value layout in the storage engine without altering Ethereum’s execution semantics. AppendChain introduces three techniques: 1) Type Separation: partitioning heterogeneous key types into dedicated column families to minimize keyspace overlap; 2) Big Value Separation: redirecting large values to external blob storage to avoid redundant rewrites during compaction; 3) Sorted Merkle Patricia Trie (MPT): applying structure-aware and order-aligned key encoding to enhance the locality of state keys. We implement AppendChain on top of Go-Ethereum (Geth) and integrate it with RocksDB, modifying the storage stack without changing Ethereum’s consensus and provenance logic. Evaluations on mainnet workloads show that, compared with Geth, AppendChain improves storage engine throughput (OPS) by up to 28 times, reduces write amplification rate nearly 94.24%, and reduces compaction frequency and time 98.75% and 72.19% respectively, significantly enhancing read/write efficiency under high-throughput transactional workloads. This framework effectively improves the storage performance of blockchain full nodes.

  • Special Issue on Data Storage and Processing for New Quality Productive Forces
  • Fanfan Shen , Zi’ang Liu , Qiwei Liang , Chao Xu , Yong Chen , Yanxiang He
    doi: 10.7544/issn1000-1239.202550483

    In the landscape of distributed intelligent systems, achieving a harmonious balance between data privacy and system efficiency remains a paramount challenge. While existing blockchain-based federated learning approaches offer enhanced security, they frequently encounter critical bottlenecks related to static resource allocation, excessive system overhead, and vulnerability to malicious participants. We propose a novel blockchain federated learning framework (FedPPB). FedPPB integrates the particle swarm optimization (PSO) for dynamic role allocation and Paillier homomorphic encryption for robust privacy preservation. The main idea of FedPPB is that PSO adaptively adjusts the number of worker, validator, and miner nodes by leveraging a multi-objective fitness function. Meanwhile, the model accuracy, verification latency, and block generation time are concurrently optimized. For security, worker nodes employ Paillier algorithm to encrypt all parameter updates before transmission. Subsequently, these parameters are decrypted and validated by validator nodes to ensure integrity. Finally, miner nodes finalize the process by generating immutable blocks to update the global model. The security of this architecture is rigorously established through theoretical analysis. Comprehensive experiments on MNIST, Fashion-MNIST, and CIFAR-10 datasets demonstrate that FedPPB significantly outperforms the existing methods, particularly in adversarial environments. Even when confronted with malicious node proportions of 15% and 25%, the proposed framework exhibits superior model accuracy and exceptional robustness, providing a sophisticated and practical solution for building secure, efficient, and adaptive decentralized intelligent systems.

  • Information Security
  • Yaxin Liu , Juan Wang , Mengda Yang , Zi’ang Li , Wenzhe Yi , Xiaoyang Xu , Meihui Chen , Tao Wei
    doi: 10.7544/issn1000-1239.202440703

    Split learning is an emerging distributed learning technique, whose main idea is to split the complete machine learning model and deploy it on the client and server respectively. During the training and inference process of the system, the client’s data are kept locally and only the encoded intermediate features are passed to the server, thus protecting the client’s data privacy to a certain extent, while alleviating the computational load of the model’s end-side operation. As split learning technology widens its application across various domains, various privacy attacks targeting split learning systems have emerged incessantly. Attackers can leverage intermediate information such as intermediate features and gradients from split layers to reconstruct users’ private data or infer their private information, posing a severe threat to data privacy. Currently, academia lacks a systematic and comprehensive overview of research achievements in split learning, with some studies confusing it with federated learning technology or offering insufficiently detailed summaries. Therefore, this paper aims to fill this gap by comprehensively summarizing relevant attack and defense techniques in split learning, providing guidance for subsequent research and development. Firstly, we introduce the definition of split learning technology, its training and inference processes, and outline its various extended architectures. Subsequently, we analyze the threat model of split learning systems and summarize the fundamental concepts, implementation stages, and existing schemes of reconstruction attacks, as well as inference attacks such as attribute inference, membership inference, and label inference targeting split learning systems. Furthermore, we summarize corresponding defense techniques, encompassing methods like anomaly detection, regularization defense, noise addition, adversarial representation training, and feature pruning. Finally, we discuss research challenges and future directions in addressing privacy and security issues in split learning.

  • Information Security
  • Zengwen Yu , Jiawei Zhang , Qi Jiang , Li Yang , Zhi Fang , Runfang Shi
    doi: 10.7544/issn1000-1239.202550411

    Dataspace is an important infrastructure for achieving effective data sharing and circulation. However, problems such as privacy leakage, data theft, and illegal abuse during the data sharing process bring big challenges to the implementation of industry dataspaces. Attribute-based encryption (ABE) can ensure the confidentiality and fine-grained access control of shared data, but there are still many problems when directly applied to dataspaces. For example, traditional ABE schemes have high computational overhead in user revocation, which cannot meet performance requirements if a lot of dynamic users join or leave the dataspace. Besides, many industry dataspaces need to exert flexible access control on shared data based on user attribute comparison and access time, and verify the decryption results. To overcome these problems, we propose a data access control approach based on comparable attributes in dataspaces, which realizes flexible and efficient user revocation to ensure forward security. The approach can make flexible decisions on access behavior based on access time and attribute comparison, and also support verification of the decryption process. Through formal security analysis, the approach has semantic security under chosen plaintext attacks. In addition, a large number of experimental analyses show that the approach is suitable for actual data sharing applications in dataspaces.

  • Information Security
  • Yaguan Qian , Qinqin Yu , Jiaqiang Sha , Qiqi Bao , Zhaoquan Gu , Shouling Ji , Bin Wang
    doi: 10.7544/issn1000-1239.202550024

    Deep neural networks (DNNs) have demonstrated remarkable success in numerous application areas, including computer vision, natural language processing, and speech recognition. However, recent studies have revealed that DNNs are vulnerable to adversarial attacks, particularly targeted attacks capable of precisely manipulating the outputs of unknown models, which poses significant risks to data privacy, model trustworthiness, and system security. Generative attack methods, which can efficiently create adversarial examples, have become a critical tool in advancing targeted attack techniques due to their ability to automate the generation process and reduce manual effort. Despite their potential, most existing generative attack approaches mainly focus on crafting adversarial examples for a single target class. This design results in clear limitations, such as low efficiency, restricted flexibility, and poor scalability in multi-target tasks, making them difficult to apply in complex scenarios where multiple targets must be addressed simultaneously. To address these challenges, we propose a multi-target generative attack based on dual-information (MTGA-DI). MTGA-DI employs a conditional generative model that integrates both semantic and visual information from multiple target classes, enabling efficient and adaptable multi-target attacks. Furthermore, the method significantly improves the transferability and stability of generated adversarial examples across different models and defense settings. Experimental results demonstrate that MTGA-DI surpasses previous methods on standard models as well as models enhanced with robust training and input preprocessing defenses, achieving higher attack success rates and better generalization.

  • Information Security
  • Yeqian Yang , Hongjun Dai
    doi: 10.7544/issn1000-1239.202440915

    The RISC-V SBI (supervisor binary interface) specification establishes a standardized interface between RISC-V hardware architectures and operating systems. In comparison to traditional firmware implementations for x86 and ARM platforms, the RISC-V SBI firmware demonstrates superior compliance with architectural standards and offers enhanced reliability. However, the current SBI implementations still exhibit significant security concerns, particularly regarding unverified boot process correctness and vulnerabilities in PMP (physical memory protection) mechanisms for physical address access control. To this end, a closed-loop formal verification framework driven by theorem proving is proposed, developing the first formally verified SBI firmware—SeSBI. By integrating Dafny and Isabelle/HOL, a layered strategy is employed to formally verify specification properties across startup process. And to improve verification efficiency, a RISC-V hardware simulation strategy is introduced, unifying the modeling of firmware-hardware interactions, thereby significantly reducing verification overhead. Finally, the functional correctness of SeSBI in the boot process is successfully verified. The results demonstrate that SeSBI implements the SBI standard boot interface with more concise code and high reliability, providing the first formally verified, trusted SBI firmware implementation for the RISC-V ecosystem. Future work will expand verification coverage to more advanced properties and optimize tooling for greater efficiency.

  • High Performance Computing
  • Wenyuan Shen , Lingtao Tang , Haotian He , Tiancheng Fei , Yulong Zhao , Lufei Zhang , Xiaojun Qin , Xin Liu , Zuoning Chen
    doi: 10.7544/issn1000-1239.202550153

    Transformer has fundamentally revolutionized various fields of natural language processing through its self-attention mechanism that effectively models long-range contextual relationships. It overcomes traditional neural networks’ limitations in handling long sequences, and has emerged as the mainstream architecture for modern large language models. Currently, Transformer inference systems generally face a shortage of computing power and are commonly confronted with challenges like substantial scheduling latency, concurrency execution difficulties, and severe memory access limitations during deployment. The Sunway AI accelerator card, featuring a domestically-developed next-generation intelligent processor, has successfully empowered diverse edge inference deployments. However, conventional inference frameworks struggle to effectively explore the architectural potential in three critical dimensions: robust system control capabilities, multi-level parallel computing, and rich memory hierarchy. We propose SW-Inf: a high-efficiency Transformer inference method based on the Sunway AI accelerator card. First, an on-chip management processing element (MPE)-orchestrated computation graph scheduling mechanism is proposed to minimize kernel launch latency and inter-process data transfer overhead through graph aggregation and MPE coordination. Second, an LDM-prioritized heuristic tiling strategy is introduced to enable load-balanced parallel computation across processing elements. Finally, a hardware-aware tightly-coupled attention computation implementation is developed to eliminate redundant intermediate data movement between registers and main memory. We implement an end-to-end inference system on the Sunway AI accelerator card. Experimental results demonstrate that compared with the vLLM inference system on NVIDIA V100S which has higher memory bandwidth than the Sunway AI accelerator card, SW-Inf achieves 15% higher throughput and 20% better memory utilization rate.

  • High Performance Computing
  • Yue Shen , Jie Zhang , Yuan Zhang , Huawei Cao , Xuejun An , Xiaochun Ye
    doi: 10.7544/issn1000-1239.202550151

    Streaming graphs can effectively model rapidly changing data in real life, and have been widely used in social network analysis, content recommendation, anomaly detection, and other fields. Based on the fact that the two graph snapshots before and after the graph update have a large amount of the same data, incremental computation reduces the amount of memory accesses and computation in the iterations by storing and reusing the historical results, which effectively improves the performance of streaming graph processing. However, the existing solutions for incremental computation optimization of graph algorithms are often limited by graph algorithms that satisfy specific properties, thus it is difficult to apply them to general graph algorithms. In this paper, the incremental computation is further divided into two modes of refinement and recomputation. The similarities and differences between the two modes are analyzed theoretically, and the performance differences between them are tested experimentally with different datasets, algorithms, and update settings. Based on the two modes, the hybrid incremental computation mode is designed to ensure the correctness of switching, and the switching time is predicted accurately by random forest classifiers. Through performance evaluation and switching effect analysis, it is proved that the general graph algorithms can effectively switch under the hybrid mode. Compared with the state-of-the-art solution DZIG, the average speedup is 1.25.

  • Artificial Intelligence
  • Qinghua Zhang , Xiong Zhou , Wei Liao , Shuaishuai Huang , Xuting Qin
    doi: 10.7544/issn1000-1239.202440289

    Joint entity and relation extraction is a foundational task for knowledge graph construction, which aims at extracting relational triples from unstructured text. To address the issues of the redundant annotations in corresponding matrix and insufficient interactions between subjects and objects, a joint model based on biaffine entity pairing and cascade tagging scheme is proposed. First, the potential relations in the sentence are predicted by performing a multi-label classification task, which eliminates the redundant relations in the relation-specific entity identification module. Then, the integrated candidate entity representations are passed through a biaffine network to enhance the interactions between subjects and objects, resulting in an entity pairing matrix that contains only candidate entities and reduces redundant annotations in the entity pairing phase. Next, a cascade tagging scheme is applied to identify relation-specific entities, and the identified entities are combined with the entity pairing matrix to form relational triples. Finally, the effectiveness of the proposed model is validated by comparative experiments and ablation study conducted on four public datasets. The experimental results demonstrate the proposed model achieves significant performance improvements over current baseline methods in terms of standard evaluation metrics, while effectively mitigating both the redundant annotation problem and the issue of insufficient entity interactions.