TeamDL at SemEval-2018 Task 8: Cybersecurity Text Analysis using Convolutional Neural Network and Conditional Random Fields

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 868–873 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics TeamDL at SemEval-2018 Task 8: Cybersecurity Text Analysis using Convolutional Neural Network and Conditional Random Fields Manikandan R 1∗, Krishna Madgula2, Snehanshu Saha1,2 1CAMMS, Dept of CSE ,PESIT-Bangalore South Campus 2PESIT-Bangalore South Campus manikandan.ravikiran@gmail.com krishnac.madgula@gmail.com snehangshusaha@gmail.com Abstract In this paper we present our participation to SemEval-2018 Task 8 subtasks 1 & 2 respec- tively. We developed Convolution Neural Net- work system for malware sentence classifi- cation (subtask 1) and Conditional Random Fields system for malware token label pre- diction (subtask 2). We experimented with couple of word embedding strategies, fea- ture sets and achieved competitive perfor- mance across the two subtasks. Code is made available at https://bitbucket.org/ vishnumani2009/securenlp 1 Introduction Cybersecurity risks and malware threats are be- coming common and increasingly dangerous re- quiring analysis of large repositories of malware related information in realtime to understand its capabilities and mount an effective defense. The sheer volume of data and its potential applica- tions alone have increased traction in recent times among NLP researchers. In this line, SemEval 2018 Task-8 offers 4 subtasks addressing text clas- sification and token, relation and attribute label prediction in cybersecurity domain using Mal- wareTextDB (Lim et al., 2017). While subtask 1 focuses on predicting sentences relevance to mal- ware , subtasks 2, 3 and 4 focus on predicting to- ken, relation and attribute labels for malware text from subtask 1. More details about the each of the subtasks can be found in Phandi et al. (2018). Concerning subtask 1, which was inherently formulated as a text classification problem very few works are done till date in cybersecurity do- main (Lim et al., 2017; Zhang et al., 2016). How- ever, in general domain the problem of text clas- sification is well addressed with extensive usage ∗Work performed during weekend part time assistantship at CAMMS of deep learning approaches (Zhou et al., 2016; Liang and Zhang, 2016; Kim, 2014; Kalchbrenner et al., 2014; Zhang et al., 2015), Support vector machines, logistic regression (Genkin et al., 2007; Jiang et al., 2016) and Tree based approaches (Bouaziz et al., 2014). On the other hand, sub- task 2 was formulated as sequence tagging prob- lem which is addressed till date by CRF (Finkel et al., 2005; R. et al., 2016, 2017), deep learn- ing approaches (Chiu and Nichols, 2016; Ma and Hovy, 2016; Lample et al., 2016) and SVM (Ekbal and Bandyopadhyay, 2012). In this paper, we describe our system that ad- dresses subtasks 1 and 2 involving malware sen- tence classification and malware token label pre- diction. We designed these systems by adapting various insights from previous works on text clas- sification and sequence tagging. We submitted a Convolutional Neural Network(CNN) based sys- tem based system for subtask 1 and Conditional Random Field (CRF) based system for subtask 2. The rest of the paper is organized as follows. In section section 2, we discuss datasets and …

Content truncated to 500 words