Shedding New Light on the Language of the Dark Web

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5621 - 5637 July 10-15, 2022 ©2022 Association for Computational Linguistics Shedding New Light on the Language of the Dark Web Youngjin Jin1 Eugene Jang2 Yongjae Lee2 Seungwon Shin1 Jin-Woo Chung2∗ 1KAIST, Daejeon, South Korea 2S2W Inc., Seongnam, South Korea 1{ijinjin,claude}@kaist.ac.kr 2{genesith,lee,jwchung}@s2w.inc Abstract The hidden nature and the limited accessibil- ity of the Dark Web, combined with the lack of public datasets in this domain, make it dif- ficult to study its inherent characteristics such as linguistic properties. Previous works on text classification of Dark Web domain have sug- gested that the use of deep neural models may be ineffective, potentially due to the linguis- tic differences between the Dark and Surface Webs. However, not much work has been done to uncover the linguistic characteristics of the Dark Web. This paper introduces CoDA, a publicly available Dark Web dataset consist- ing of 10000 web documents tailored towards text-based Dark Web analysis. By leveraging CoDA, we conduct a thorough linguistic anal- ysis of the Dark Web and examine the textual differences between the Dark Web and the Sur- face Web. We also assess the performance of various methods of Dark Web page classifica- tion. Finally, we compare CoDA with an exist- ing public Dark Web dataset and evaluate their suitability for various use cases. 1 Introduction The World Wide Web contains a vast, non-indexed part of the Internet (known as the Deep Web) which is hidden from traditional web search engines. The Dark Web, which refers to the small portion of the non-indexed pages that require specific routing protocols such as Tor1 for access, has become a safe haven for users wanting to conceal their identity and preserve their anonymity. A consequence of the properties of the Dark Web (limited methods of access and the volatility of its onion services) is that it is difficult to grasp the general topology and the overall content of the Dark Web. A number of past academic studies have tried to unravel the Dark Web through methods such as page classification (Al Nabki et al., 2017; ∗Corresponding author 1Tor Project: https://www.torproject.org/ Ghosh et al., 2017; He et al., 2019; Choshen et al., 2019) and content analysis (Biryukov et al., 2014; Avarikioti et al., 2018). However, not much work has been done on the linguistic analysis of the Dark Web (Choshen et al., 2019). In addition, the Dark Web has been studied and analyzed in the security research community to uncover malicious activities including phishing (Yoon et al., 2019), illicit online marketplace ac- tivity (Soska and Christin, 2015), terrorism (Chen, 2011), cryptocurrency abuse (Lee et al., 2019), and ransomware ecosystems (Meland et al., 2020). We believe that the lack of a comprehensive work on the language of the Dark Web from the NLP com- munity mainly stems from the lack of Dark Web datasets publicly available for research. Therefore, a new dataset on the Dark Web …

Content truncated to 500 words