Title: Leveraging NLP and Social Network Analytic Techniques to Detect Censored Keywords: System Design and Experiments
Internet regulation in the form of online censorship and Internet shutdowns have been increasing over recent years. This paper presents a natural language processing (NLP) application for performing cross country probing that conceals the exact location of the originating request. A detailed discussion of the application aims to stimulate further investigation into new methods for measuring and quantifying Internet censorship practices around the world. In addition, results from two experiments involving search engine queries of banned keywords demonstrates censorship practices vary across different search engines. These results suggest opportunities for developing circumvention technologies that enable open and free access to information.  more » « less
Award ID(s):
1704113
PAR ID:
10463464
Author(s) / Creator(s):
;
Date Published:
Journal Name:
52nd Hawaii International Conference on System Sciences (HICSS)
Subject(s) / Keyword(s):
Sponsoring Org:
National Science Foundation
More Like this
  1. The commoditization of deep packet inspection technology has eased the deployment of Internet censorship, even in countries typically considered open. While the Internet freedom community has been vital to increasing transparency, its coverage relies heavily on NGOs and activists, whose ability to relay user reports of censorship is increasingly threatened by mounting risks, anti-NGO legislation, and foreign funding cuts. To bridge this widening gap, we examine whether other data sources can be repurposed to amplify user voices and rally censorship investigation. We leverage Google Trends data to build CenAlert, a user-driven alert system that detects spikes in search interest for circumvention tools. Each spike is scored with an impact factor quantifying increases in circumvention tool demand and enabling prioritization of response efforts. We demonstrate the effectiveness and practicality of CenAlert across 76 censoring countries over 14 years. Of 269 selected spikes, 191 are explainable, including 153 as censorship events. Notably, 68 spikes correspond to censorship events not previously reported by the community. To facilitate integration with observatories, CenAlert is open-source and provides a dashboard, API, and Slack webhook for notification, visualization, and analysis of potential censorship events. Ultimately, CenAlert offers a novel solution to long-standing challenges faced by the Internet freedom community. 
    more » « less
  2. Internet censorship is pervasive, with significant effort dedicated to understanding what is censored, and where. Prior censorship measurements however have identified significant inconsistencies in their results; experiments show unexplained non-deterministic behaviors thought to be caused by censor load, end-host geographic diversity, or incomplete censorship—inconsistencies which impede reliable, repeatable and correct understanding of global censorship. In this work we investigate the extent to which Equal-cost Multi-path (ECMP) routing is the cause for these inconsistencies, developing methods to measure and compensate for them. We find that ECMP routing significantly changes observed censorship across protocols, censor mechanisms, and in 18 countries. We identify that previously observed non-determinism or regional variations are attributable to measurements between fixed endhosts taking different routes based on Flow-ID; i.e., choice of intrasubnet source IP or ephemeral source port leads to differences in observed censorship. To achieve this we develop new route-stable censorship measurement methods that allow consistent measurement of DNS, HTTP, and HTTPS censorship. We find ECMP routing yields censorship changes across 42% of IPs and 51% of ASes, but that impact is not uniform. We develop an application-level traceroute tool to construct network paths using specific censored packets, leading us to identify numerous causes of the behavior, ranging from likely failed infrastructure, to routes to the same end-host taking geographically diverse paths which experience differences in censorship en-route. Finally, we compare our results to prior global measurements, demonstrating prior studies were possibly impacted by this phenomenon, and that specific results are explainable by ECMP routing. Our work points to methods for improving future studies, reducing inconsistencies and increasing repeatability 
    more » « less
  3. Internet censorship imposes restrictions on what information can be publicized or viewed on the Internet. According to Freedom House’s annual Freedom on the Net report, more than half the world’s Internet users now live in a place where the Internet is censored or restricted. China has built the world’s most extensive and sophisticated online censorship system. In this paper, we describe a new corpus of censored and uncensored social media tweets from a Chinese microblogging website, Sina Weibo, collected by tracking posts that mention ‘sensitive’ topics or authored by ‘sensitive’ users. We use this corpus to build a neural network classifier to predict censorship. Our model performs with a 88.50% accuracy using only linguistic features. We discuss these features in detail and hypothesize that they could potentially be used for censorship circumvention. 
    more » « less
  4. Active Internet measurements are crucial for exposing the increasing frequency and severity of global Internet censorship. However, current measurement efforts are constrained by limited resources and time, and reacting to new censorship events remains a largely manual process reliant on rapid signals. As a result, creating a comprehensive and real-time picture of Internet censorship remains a key challenge. In this work, we introduce CenRL, an intelligent censorship measurement framework that leverages reinforcement learning to optimize and automate censorship measurements. We model the censorship measurement process as a multi-armed bandit problem and design CenRL agents to address two key tasks: maximizing the detection of blocked websites within a network and automatically responding to blocking changes in dynamic censorship environments. We demonstrate CenRL’s effectiveness through realistic simulated experiments in three highly censored regions (China, Russia, and Kazakhstan) and real-world censorship measurements across vantage points in 15 countries with diverse censorship policies. Our controlled experiments demonstrate that CenRL significantly outperforms the state of the art measurement processes, finding 75% of blocked websites in less than half the number of measurements while identifying censorship changes up to seven times faster. Our real-world experiments confirm this advantage across multiple censorship environments, showing that CenRL can find 2.5 times more blocked websites on average compared to existing measurement strategies. Our study demonstrates the potential of using reinforcement learning to provide deeper insights into restrictions on online freedom. 
    more » « less
  5. If it remains debatable whether the Internet has surpassed print media in making information accessible to the public, then it must nevertheless be conceded that the Internet makes the manipulation and censorship of information easier than had been on the printed page. In coming years and in an increasing number of countries, everyday producers and consumers of online information will likely have to cultivate a sense of censorship. It behooves the online community to learn how to detect and evade interference by governments, regimes, corporations, con-artists, and vandals. The contribution of this research is to describe a method and platform to study Internet censorship detection and evasion. This paper presents the concepts, initial theories, and future work. 
    more » « less