![](bg6.png)
4.2 Experimental Parameter Setting
This paper embeds the 64-dimensional word2vector
vector initialization word, and all weight parameters
are initialized uniformly. Adam is selected as the
optimizer for the hidden layer with a dimension of 64.
Use a learning rate of 0.001, a random inactivation
parameter of 0.01, and a batch size of 256. A random
initialization of 100 times was performed, and the
results of the average of those 100 times were taken
as the final results. Multiple dimensions are used to
evaluate the effectiveness and robustness of the
experimental results, including accuracy, recall, and
F1 value.
4.3 Experimental Structure Analysis
In Table 2, experimental results demonstrate that the
performance of this method is superior to that of other
comparable models, providing evidence for its
effectiveness. Using two network models to capture
semantic data from multiple directions, this paper
successfully reduces the gradient of model training
data by analyzing the SwissLog log and separating
the log workflow. This also improves the training
effect and efficiency, and eliminates the problem of
overfitting.
Table 4. Comparison of experimental effects
5 CONCLUSION
Several studies have been conducted on the text
classification of traditional texts, but little research
has been conducted on the text classification of
update logs. Update logs contain a great deal of
functional and security information, so they are
valuable for future security research and topic
annotation. In order to improve SwissLog's parsing
logs based on the characteristics of updating logs, we
propose a new TransformerGRU network model and
develop a new workflow for log serialization
preprocessing that is more efficient and accurate.
Furthermore, the self-attention model is used to
balance different log key data sets with a high degree
of quality, which further enhances the classification
effect. Experiments have demonstrated that this
method is highly accurate and efficient for classifying
objects.
ACKNOWLEDGEMENTS
This project is supported by Shan dong Province
Science and Technology Small and Medium
Enterprises Innovation Ability Enhancement Project
of China (No. 2023TSGC0449)
REFERENCES
X. Li, P. Chen, L. Jing, Z. He and G. Yu, "SwissLog:
Robust and Unified Deep Learning Based Log Anomaly
Detection for Diverse Faults", 2020 IEEE 31st
International Symposium on Software Reliability
Engineering (ISSRE), Coimbra, Portugal, 2020, pp. 92-
103.
Li, Xiaoyun and Chen, Pengfei and Jing, Linxiao and He,
Zilong and Yu, Guangba, "SwissLog: Robust Anomaly
Detection and Localization for Interleaved Unstructured
Logs", 2022 IEEE Transactions on Dependable and
Secure Computing, Coimbra, Portugal, 2022.
Sakhnini, S., & Alazab, M. (2019). Rule-Based Anomaly
Detection Algorithm for Web Server Log Files. IEEE
Access, 7, 24478-24486.
Hou, M., Jin, R., & Zhou, S. (2019). A statistical approach
for anomaly detection in web log data. Journal of
Network and Computer Applications, 131, 52-59.
N. Feng, Y. Zhang, Y. Chen, Detecting Anomalies in Large
Scale Logs via Deep Reinforcement Learning. In IEEE
Transactions on Information Forensics and Security,
vol. 15, pp. 2020,1185-1198.
L. Forgor, W. Brown-Acquaye, J. K. Arthur and S. Owoo,
"Security of Data on E-waste equipment to Africa: The
Case of Ghana, " 2019 International Conference on
Communications, Signal Processing and Networks
(ICCSPN), Accra, Ghana, 2019, pp. 1-5.
Z. Liu et al., "Swin Transformer: Hierarchical Vision
Transformer using Shifted Windows," 2021 IEEE/CVF
International Conference on Computer Vision (ICCV),
Montreal, QC, Canada, 2021, pp. 9992-10002.
Transformer-Encoder-GRU (T-E-GRU) for Chinese
Sentiment Analysis on Chinese Comment Text by
Binlong Zhang, Wei Zhou, 2021, 211-234.
Fuhong Tang and Kwankamol Nongpong. Chinese
sentiment analysis based on lightweight character-level
bert. In 2021 13th International Conference on Kno
wledge and Smart Technology (KST), pages, 2021 27–
32.
T. Kongsin and S. Klongboonjit, "Machine Component
Clustering with Mixing Technique of DSM, Jaccard
Distance Coefficient and k-Means Algorithm" 2020
IEEE 7th International Conference on Industrial
Engineering and Applications (ICIEA), Bangkok,
Thailand, 2020, pp. 251-255.
O. Ertl, "ProbMinHash – A Class of Locality-Sensitive
Hash Algorithms for the (Probability) Jaccard
Similarity, " in IEEE Transactions on Knowledge and
Data Engineering, vol. 34, no. 7, 2022, pp.3491-3506.
ANIT 2023 - The International Seminar on Artificial Intelligence, Networking and Information Technology
420