Design of Enterprise Data Security System Based on Lightweight LSTM-CRF
DOI: https://doi.org/10.62517/jbdc.202501314
Author(s)
Lan Li, Jinlong Tan, Fabin Yang, Jundi Wang*
Affiliation(s)
School of Computer and Artificial Intelligence, Lanzhou Institute of Technology, Lanzhou, Gansu, China
*Corresponding Author
Abstract
In enterprise digitalization, enterprise data security must tackle both internal operational risks and emerging external attacks like Fuzzers Malformed Packet Injection. As a classic network intrusion detection benchmark dataset, NSL-KDD’s preprocessing and lightweight model training are critical for enterprise scenario adaptation. However, small and medium-sized enterprises (SMEs) face a prominent contradiction: limited resources versus urgent security needs. Traditional rule engines rely on manual updates, leading to low unknown threat recognition rates and high false positives; existing deep models such as LSTM and BERT have massive parameters and high resource consumption, hard to deploy on SMEs’ CPUs; standard LSTM-CRF is also plagued by sample imbalance, resulting in weak generalization. To solve these issues, this study designs a three-layer collaborative system centered on Lightweight LSTM-CRF, supported by multimodal log preprocessing, self-evolving defense and blockchain audit algorithms, with Dynamic Trust Evaluation embedded. Experiments on the NSL-KDD dataset in a common CPU environment show the system achieves over 90% anomaly recognition accuracy, less than 100ms threat response time, and 55% lower deployment costs, accurately meeting SMEs’ needs.
Keywords
Lightweight LSTM-CRF; Enterprise Data Security; Dynamic Trust Evaluation; System Design
References
[1]Libin Xie, Yan Zhang, Lili Zhou. Analysis and prospect of data security risk situation and governance trends. China Telecommunication Industry, 2025, (07): 61-63.
[2]Verizon. 2024 Data Breach Investigations Report. United States: Verizon, 2024.
[3]Xianlong Zhang. Thoughts on enterprise data security protection in the digital economy era. Digital Communication World, 2022, (02): 116-118.
[4]Ru Li, Shuying Zhai, Bo Li. Research on network security defense system for small and medium-sized enterprises. Microcomputer Applications, 2023, 39 (06): 1-3.
[5]Yunan Zhang, Chao Hong, Yiwei Yang, Pandeng Li, Xiaoyun Kuang, Yixin Jiang. Research on a new architecture of identity authentication and authorization based on zero trust network security. Network Security Technology and Application, 2025, (07): 19-23.
[6]Jutao Shan, Lihong Guo, Yufei Ji, et al. Network intrusion detection system based on deep learning. Internet of Things Technology, 2025, 15 (05): 63-67.
[7]Lianhai Liu, Huiye Li, Donghui Mao. CBAM-CNN network intrusion detection method based on image convex hull features. Information Network Security, 2024, 24 (09): 1422-1431.
[8]Gang Wang, Qian Peng, Hongjun Duan, Wenhua He. Design and implementation of computer network security defense system based on artificial intelligence technology. Heilongjiang Science, 2024, 15 (18): 70-73.
[9]Cen Chen, Zhihao Qu, Ming Wang, et al. Zero trust dynamic access control for power grid security. Journal of Chongqing University, 2024, 47 (08): 81-89.
[10]Xin Tang, Xiongwei He, Xiangjie Lü. Research on optimization algorithm of mIBSG scheme based on blockchain. Telecommunications Technology, 2025, (06): 86-89.