舆情秘书 : Do you hear the people sing ?
An investigation into a mass online surveillance tool sought after by several Chinese government security agencies.
Previously, we ran into an digital tool called 舆情秘书 (“Public Opinion Secretary”). Designed for mass data analysis of social media, websites and other online venues where humans express and exchange opinions, ideas and concerns.
The tool is openly advertised as a "Big data service” by its maker, Beijing Wisdom Starlight Information Technology on their website. Comparable maybe to international products from Dataminr, Meltwater or Pulsar Platform. The list of use cases on their website, ranges from government agencies to data research and economical profiling. There is even a "free” version on the usual app stores.
The brochure lists the target markets the company is trying to sell their product to. The order is not a coincidence I guess:
行业一：外事机构 (Foreign Affairs Agencies/Institutions )
行业二：党政机关、事业单位 (Party and government organs, public institutions)
行业三：国企社团 (State-owned enterprises)
行业四：商业企业 (Commercial enterprises)
But public procurement data from China shows that it is the Chinese Ministry of Public Security (MPS) and the Ministry of State Security (MSS) that seem increasingly interested in using it. So in that function it takes potentially a lot of inspiration from Palantir’s Gotham program.
We can find at least five different Public Security Bureaus in China having this tool on their wish list plus the tender by the Ministry of State Security mentioned in an earlier post.
Further, the tools seems actively marketed to military and security services as seen via this announcement for the upcoming China Military Intelligent Technology Expo: “the Public Opinion Secretary can achieve real-time collection, in-depth analysis and accurate early warning of public opinion data at home and abroad, helping the military and public security departments to accurately grasp the public opinion trends.”
Despite not available on their website an official sales brochure in Chinese could be unearthed from the more obscure corners of the Internet. Though it seems from an earlier version, around 2017 most likely, this document will be our guide to delve deeper into the functionality of this product.
The product itself is developed and sold by “Beijing Wisdom Starlight Information Technology” corporation. A Beijing based technology company that is often involved in providing tools and services to Chinese government entities. The company has deep connections, including overlapping management personal, with Fujian Rear Information Technology and Southeast Public Opinion Datacenter. Companies that often appear on tender documents of other government security agencies in China as well. It is also a official Amazon Web Services (AWS) partner.
Of course, this document is a sales brochure. Therefore you can expect some glorified description of the products capabilities but it serves as well as a good guide to explore the functionality.
The brochure defines the software in the following words: "舆情秘书" is a precision-grade surveillance and public opinion management system used by government, SOEs, and education institutions in China. It combines AI, big data, and real-time notification to identify, assess, and defuse sensitive or negative narratives — especially politically charged or high-impact events.
Features
Let's take a look at the list of functions and features in the document:
Full-spectrum monitoring: Tracks news, blogs, forums, WeChat, Weibo, video platforms (e.g., Douyin, Bilibili), e-commerce reviews, and even overseas sites (Facebook, Twitter).
Image OCR: Detects and extracts text from images posted online, such as signs, banners, or screenshots.
Event analysis: Clusters posts into events, traces sources, analyzes dissemination path, and identifies key influencers.
Sentiment analysis: Classifies content as positive, neutral, or negative with up to 97% accuracy, using AI-powered language processing.
Real-time alerts: Pushes keyword and sentiment-triggered alerts to WeChat, email, or PC popups.
Short video monitoring: Includes platforms like TikTok (Douyin), Kuaishou, Xigua, etc.
Social monitoring: Monitors QQ and WeChat group chats.
TV monitoring: Monitors TV news channels for specified keywords.
Report generation: Produces daily/weekly/monthly visualized reports, custom summaries, and expert briefs.
Interesting enough it explicitly mentions Facebook and Twitter, two services that are officially not accessible from Mainland China.
Machine heart
One main element of this software heavily advertised in this document is it's ML / AI functionality to achieve and extend this feature set. Lets walk through some of the key expressions we can find in the brochure:
机器智能预警 + 人工审核，信息更精准: Suggests use of automated classification models for triaging or flagging sensitive content, likely powered by supervised machine learning.
仿照人类思维逻辑，实现人工智能逻辑判别处理，负面信息判负比例高达97%: Implies the use of AI models to detect and label sentiment in public discourse, possibly using NLP classification.
自然语义识别、去噪、聚类、建立模型，实现智能分析: A classic machine learning pipeline — unstructured text is processed and clustered to detect emerging topics or incidents.
高级逻辑运算，智能情感识别，准确率高达97%: Strong implication of sentiment analysis models, often based on ML or deep learning.
支持自动生成日报…信息筛选，一键生成可视化简报: Suggests automated reports and formatting, which may include ML-based content selection.
How it works - an example
The brochure also gives us a general walk through of how the tool works in a real life scenario.
The base premise was that a viral video appeared, showing a drunken judge in a provincial court for a hearing. It attracted quite some online comments, questioning the quality of the judicial system in the region.
The brochure walks the reader through the steps that the software went through to help to mitigate the issue.
1. Viral Event Detection
The system automatically detected the viral video based on:
Keywords like “法官” (judge), “醉酒” (drunk), “法庭” (court)
Real-time monitoring of Weibo, forums, and video platforms
舆情秘书 triggered an alert within minutes of the video being posted, both on the PC dashboard and via mobile app push notification.
2. Internal Notification and Mobilization
Leadership and public information officers were immediately notified via:
Mobile phone alerts (push + link)
Email notifications
The court activated its emergency response team and began investigating the incident.
3. Event Verification and Damage Control
Officials used the system to:
Analyze the source of the video
Trace how it spread across platforms
Identify influential accounts amplifying the event
Using this insight, they verified the video’s authenticity and pinpointed the involved judge.
4. Crisis Response Decision
Based on the system’s data and automatically generated event report, the leadership decided to:
Suspend the judge
Announce an internal investigation
Issue a public statement through official channels
5. Outcome
The rapid response and transparent handling helped to defuse public anger.
The system tracked sentiment after the official response and showed a decline in negative sentiment.
Eventually the company claims that “Because of the prompt response, the wave of online negative sentiment quickly subsided.”
Of course such statements are hard to verify. But the circumstance that several requests by public security institutions to acquire this specific tool, gives the impression that this is not just hollow “marketing talk”.
Another example on the official webpage mentions a scenario in which students of a Chinese University posts information on foreign websites about internal university happenings that can't be controlled by the local propaganda department and therefore cause damage to the reputation of the University and the nation. The website recommends the use of Public Opinion Secretary to catch such information leaks, alert the responsible parties and provide them with options to remedy the situation or counter the fallout.
Conclusion
None of this is secret information of course. Technically, it's just another big data analysis tool—one of many on the open market. Offered as a solution for a very broad set of challenges. Even in China, there are several competitors. But, as with many things in China, context and use-case is what ultimately matters.
We see the proliferation of such big data analysis tools in law enforcement and other government agencies around the world. Most notably in the United States, companies like Palantir and Dataminr frequently come up, often in discussions about fears that governments might overstep their bounds in pursuit of social control.
In China, of course, such debates are rare, nor are they widely publicized. The government’s grip on society has long been tighter than in many Western democracies, so it’s no great surprise to many that the new capabilities of mass data analysis, real-time monitoring, and the rise of machine learning are opening entirely new avenues for surveillance.
From my own experience, these developments have become increasingly noticeable in recent years—likely accelerated by pandemic-control-related measures. Online spaces where people could once express their views relatively freely are now either gone or being sanitized more rapidly than ever. It’s a gradual shift, but a visible one—if you're willing to look up and pay attention.
The system still has holes, and it remains surprising how much slips through when the infrastructure is overwhelmed. When pressure mounts, the sheer volume of information flooding the Chinese internet becomes difficult to manage—even for one of the world’s most sophisticated surveillance states.
But tools like 舆情秘书 (Public Opinion Secretary) offer a glimpse into the direction this is heading and the focus of China’s surveillance industry in general. Like their Western counterparts, these tools work with citizens’ growing habit of sharing their lives online. Despite these organizations wielding immense surveillance powers over even private aspects of digital communications between citizens in China, the need to keep an close eye on the public discourse across all sorts of media is very important to feel the pulse of society. The difference is that in China, once national security institutions get involved, the guardrails—if they ever existed—are off. Lofty talk about protecting consumer data from corporate misuse and mass data collection for commercial purpose sounds hollow, when the state itself is very much interested in hoovering up every little piece of information about their citizen’s online life to profile for the purpose of mass control and surveillance.