Enrichment of Features for Malware-Related Sentence Classification using External Knowledge
Enrichment of Features for Malware-Related Sentence Classification using External Knowledge
复制标题
DOI：
10.1109/ictai52525.2021.00181
复制
发表时间：
2021-11
期刊：
影响因子：
--
通讯作者：
Chau Nguyen;Vu Tran;Minh Le Nguyen
中科院分区：
文献类型：
--
作者：
Chau Nguyen;Vu Tran;Minh Le Nguyen
识别与恶意软件相关的句子是我们使用自然语言处理技术对恶意软件报告进行更深入分析的第一步。然而，这种二进制分类任务对于神经网络方法来说甚至是具有挑战性的，因为这些方法在任务上的性能远非完美。以前的方法集中在只使用带注释的数据来处理任务，这可能会限制分类器的性能。在本文中，我们建议利用外部知识来丰富句子的特征。实验结果表明，支持向量机（SVM）模型的F1得分增加了约9%，与没有丰富的功能的模型的性能相比。我们还在识别恶意软件相关句子的任务中获得了最佳F1分数。
Identifying malware-related sentences is the first step before we can perform deeper analysis on malware reports using natural language processing techniques. However, this binary classification task is even challenging for neural network approaches as the performance of those approaches on the task is far from perfect. The previous approaches focused on using only annotated data to tackle the task, which may limit the performance of the classifiers. In this paper, we propose to leverage external knowledge to enrich the features of the sentences. The experimental results demonstrate that, with the enriched features, a support-vector machine (SVM) model gains about 9% on F1 score comparing to the performance of the model without the enriched features. We also achieve the best F1 score on the task of identifying malware-related sentences.