[Paper Review] Automatically Generating a Large, Culture-Specific Blocklist for China

Austin Hounsel, Prateek Mittal+1 more|arXiv (Cornell University)|Jun 4, 2018

Internet Traffic Analysis and Secure E-voting8 references3 citations

TL;DR

This paper presents a method to automatically generate a large, culture-specific blocklist of 1,125 censored websites in China by leveraging natural language processing (NLP) on Chinese text to extract sensitive phrases, which are then used as search queries to discover censored domains. The approach significantly expands existing blocklists by focusing on culturally relevant, non-English content, with none of the discovered sites present on the largest prior blocklist, thus enabling more comprehensive censorship measurement.

ABSTRACT

Internet censorship measurements rely on lists of websites to be tested, or "block lists" that are curated by third parties. Unfortunately, many of these lists are not public, and those that are tend to focus on a small group of topics, leaving other types of sites and services untested. To increase and diversify the set of sites on existing block lists, we use natural language processing and search engines to automatically discover a much wider range of websites that are censored in China. Using these techniques, we create a list of 1125 websites outside the Alexa Top 1,000 that cover Chinese politics, minority human rights organizations, oppressed religions, and more. Importantly, $ extit{none of the sites we discover are present on the current largest block list}$. The list that we develop not only vastly expands the set of sites that current Internet measurement tools can test, but it also deepens our understanding of the nature of content that is censored in China. We have released both this new block list and the code for generating it.

Motivation & Objective

• To address the lack of comprehensive, culture-specific blocklists for measuring internet censorship in China.
• To overcome the limitations of existing blocklists that focus on English content and miss censored websites in Chinese.
• To develop an automated system that discovers censored websites using NLP-processed Chinese phrases as search queries.
• To create a publicly available, large-scale blocklist that includes marginalized voices and culturally relevant content from Chinese perspectives.
• To deepen understanding of the scope and nature of content censorship in China beyond politically sensitive keywords.

Proposed method

• Extracting content-rich, sensitive Chinese phrases from existing censored web pages using NLP techniques tailored for Chinese text, which lacks word spacing.
• Using these multi-word phrases (bigrams, trigrams) as search queries in Chinese search engines like Bing to find related censored websites.
• Testing each search result for DNS manipulation in China to determine if it is blocked, using non-DNS IP addresses to detect interference.
• Iteratively feeding newly discovered censored domains back into the system to expand the search for related content.
• Applying TF-IDF to identify high-impact keywords from the Citizen Lab blocklist for initial query generation.
• Parsing Chinese text to ensure censored websites written entirely in simplified Chinese are not overlooked, unlike in prior work.

Experimental results

Research questions

• RQ1 Can NLP-processed Chinese phrases yield more effective and contextually relevant search queries for discovering censored websites in China than single-word queries?
• RQ2 To what extent do politically sensitive phrases in Chinese, such as references to historical events or government figures, correlate with website blocking in China?
• RQ3 How does the inclusion of non-English, culturally specific content improve the comprehensiveness and representativeness of censorship blocklists?
• RQ4 Can a system that uses multi-word, context-rich phrases discover censored websites not present in existing large-scale blocklists?
• RQ5 What types of content—such as human rights, religious, or political dissent—are most frequently targeted in China’s internet filtering, based on phrase-based discovery?

Key findings

• The system discovered 1,125 censored domains in China that are not present on the largest existing blocklist, significantly expanding the scope of censorship measurement.
• The blocklist is 12.5 times larger than the most widely used blocklist for censorship research, enhancing the diversity and coverage of testable websites.
• Phrases such as "1989 democracy movement" and "Tienanmen Square demonstrations" showed high block rates (30% and 32%, respectively), indicating strong correlation with censorship.
• Bigrams like "Chinese Communist Party’s religious policy" and "illegal detention" had block rates of 42% and 36%, respectively, indicating systematic targeting of political and human rights content.
• The use of multi-word Chinese phrases, such as "Chinese human rights violation," led to the discovery of culturally relevant websites not found via single-word queries, including activist and expatriate-run sites.
• The blocklist includes a wide range of censored content, including minority rights, religious freedom, political dissent, and privacy-enhancing technologies, reflecting diverse marginalized perspectives.