A Naive Bayes classifier built on 17 LIWC-derived and keyword features achieved 79.34% accuracy (10-fold cross-validation) predicting censorship of Sina Weibo posts, with precision 0.80 and recall 0.85 for the censored class — outperforming all single-domain feature sets including the full 408-feature combination (0.69 accuracy).

Information Gain feature selection from 408 candidates identified informal language markers (informal, nonflu, swear), Chinese modal and general particles signaling mood and relational framing, and physical-feeling words used metaphorically as the top predictors of censored Weibo content — all with statistically significant differences between censored and uncensored classes.

§5, §6 detection

A 598-term sensitive-keyword blacklist (sourced from Wikipedia and China Digital Times) achieved only 53% classification accuracy on Weibo censorship — below the 66% achieved by punctuation features alone — and appeared in only 31 of 152 uncensored posts versus 60 of 192 censored posts, confirming keywords are not the primary driver of platform censorship decisions.

§4.2, §6, Table 1 detection

Sentiment analysis features (Baidu and Boson tools) achieved only 57% accuracy individually; censored posts averaged 53.9% negative sentiment (General model) versus 49.3% for uncensored — a difference too small to be operationally useful — indicating that sentiment polarity does not reliably distinguish censorable content from permitted content on Weibo.

§4.3–4.5, §6, Table 1 detection

When the researchers attempted to use Gemini 2.5 Flash as a third independent LLM judge via its API for evaluating moderation decisions, Gemini automatically blocked all judging attempts citing safety reasons. This occurred even though the research task (judging whether a response is more or less moderated) does not itself produce harmful content. The incident illustrates that LLM safety systems can over-block legitimate research use cases, and that different LLM providers have different thresholds— Claude Haiku 4.5 and GPT-4o completed all judging tasks without safety refusals.

2026-lipphardt-dual §3.3.3 ml-classifierkeyword-filtering

Category-level analysis of 100 statements across 5 sensitive content categories found that interface-based moderation gaps vary significantly by topic. Sexuality showed the strongest WebUI/API gap (WebUI 7.0× more likely to be moderated than API per GPT-4o judge for Gemini). Political ideology followed at 2.0×, then hate speech at 1.0×. Miscellaneous offensive topics showed the inverse pattern (API more moderated at 0.3×). Religious content showed WebUI moderation with no API moderation. The pattern suggests public-facing WebUI interfaces prioritize reputational risk management for high-scrutiny categories.

2026-lipphardt-dual §4.5, Figure 8 ml-classifierkeyword-filtering

An empirical study of 100 sensitive statements tested on Gemini (2.5 Flash) and ChatGPT (GPT-5) found that WebUI interfaces are systematically more restrictive than their API counterparts. According to GPT-4o judge: WebUI was moderated 18% of the time vs. 9% (Gemini API) and 13% (ChatGPT API). DeBERTa classifier found 82% of WebUI responses moderated vs. 58% of API responses. The Gemini WebUI:API ratio ranged from 2.0:1 (GPT-4o) to 7.0:1 (Claude), and ChatGPT from 1.4:1 (GPT-4o) to 15.6:1 (Claude). Neither Google nor OpenAI discloses these interface-specific policies.

2026-lipphardt-dual §4.3, Table 2 ml-classifierkeyword-filtering

Across all 7 LLMs tested (GPT 4o, GPT 4o Mini, Gemini 1.5 Flash, Gemini 1.5 Pro, Llama 3.2, Claude 3.5 Haiku, Claude 3.5 Sonnet), statistically significant evidence of censorship bias was found in at least one evaluation metric per model: responses to Simplified Chinese prompts were more neutral, more similar to sanitized text, and less opinionated than semantically identical Traditional Chinese prompts (p < 0.05 across refusal-rate, sentiment, CensorshipDetector classification, and word-embedding analyses).

2025-ahmed-llm-censorship-bias Table 4, §6 keyword-filteringml-classifier

CensorshipDetector, an XLM-RoBERTa model fine-tuned on 587,819 Baidu Baike articles (censored) and Chinese Wikipedia (uncensored), achieved 91% accuracy on a held-out validation set of Chinese news articles, correctly classifying 93% of Chinese state media articles as censored and 87% of New York Times Chinese articles as uncensored, with average censorship scores of 0.93 and 0.13 respectively.

2025-ahmed-llm-censorship-bias §4.7, §4.7.2 keyword-filteringml-classifier

The paper proposes a black-box methodology for detecting censorship bias in LLMs by comparing responses to identical prompts in Simplified vs. Traditional Chinese — scripts for the same spoken language — controlling for translation quality while exploiting that Simplified Chinese training data is disproportionately sourced from mainland China's censored internet. Each prompt is repeated ten times and scored for similarity to censored text using an XLM-RoBERTa classifier fine-tuned on Baidu Baike (censored) vs. Chinese Wikipedia (uncensored) with scores from 0 to 1.

2024-ahmed-extended §2 / §2.4 keyword-filteringml-classifiermeasurement-platform