Reasons why regulatory agencies and regulated companies should prepare for online monitoring
The scale of posts related to items under the jurisdiction of the Ministry of Food and Drug Safety distributed online has exceeded the capacity for officials to manually check by entering keywords into the search bar.
The online monitoring system of the Ministry of Food and Drug Safety has also changed accordingly. It collects about 310,000 posts from various online channels such as Naver blogs, social media, and e-commerce over the past 30 days, and AI analyzes the level of risk to select about 1,000 posts for priority review.
The significance of this case lies not only in the fact that the Ministry of Food and Drug Safety has introduced crawling technology.
The point is that regulatory agencies are now collecting online posts and sellers on a large scale, managing the level of risk and frequency of exposure as data.
For public agencies and associations, this serves as a standard for how to streamline online management systems, while for companies in the food, health functional food, pharmaceutical, cosmetics, and medical device industries, it poses more direct questions.
If regulatory agencies are viewing our products and sellers' posts as data, are we checking our own and distribution network first?
Limits faced while operating 170 crawlers directly
Over the past three years, the Ministry of Food and Drug Safety has operated about 170 crawlers for its own infrastructure to find false or exaggerated advertisements and illegal distribution information of food and pharmaceuticals online.
They had defined surveillance targets and possessed the technology to collect data. The issue was not creating the crawlers but rather maintaining their normal operation over an extended period.
Operating 170 crawlers for an extended period leads to ongoing tasks such as:
- Modifying collection logic due to site structure changes
- Dealing with IP blocking, access restrictions, captcha, and login policy changes
- Checking collection errors and data omissions
- Adding new channels and keywords for collection, managing collection cycles, and server operations
The essence of surveillance work is to identify potentially violating posts for review and action. However, when personnel are tied up in operating crawlers, their tasks shift from judging violating posts to maintaining crawlers.
Companies face a similar situation. The range of items that need human verification rapidly expands to include not only official channels but also distributors, sellers, resellers, open market sellers, and influencer content.
As the number of entities to manage increases, the key is not for humans to view all posts but to establish a system for prioritizing which posts to check first.
Automate collection and focus on judgment
Online compliance monitoring involves continuously collecting online posts related to regulated products and analyzing the potential for violations to select targets that require immediate review by personnel.
The flow of the redesigned system is as follows:
- Channel-specific crawlers automatically collect new posts.
- Analyze post titles, content, items, advertising expressions, and sales situations.
- AI assigns a risk score to each post.
- Only posts with high risk are forwarded for personnel review.
- E-commerce posts are automatically aggregated by seller (business entity).
- Monitor trends of repeated high-risk sellers and channels on a dashboard.
AI does not make the final decision on legal or administrative violations. Its role is to prioritize selecting high-risk posts from a large volume of posts for review, with the final decision left to personnel.
Collected data includes post content, collection channels, item categories, matching keywords, AI risk levels, and seller matching status. This allows for not only collecting posts but also identifying which channels and sellers have a concentration of high-risk posts.
Collecting 310,000 posts monthly and prioritizing about 1,000 for review
Here is the operational status as of the past 30 days:
| Item | Operational Status |
|---|---|
| Amount collected within the period | About 310,000 posts |
| Increase in collection | Over three times compared to the previous period after expanding the collection scope |
| Surveillance channels | 7 channels |
| Monitoring keywords | 100 |
| AI high-risk classification | About 1,000 posts |
| Automatically aggregated e-commerce sellers | Over 2,800 |
| Collection frequency | Every 3 hours or twice a day depending on the channel |
The most significant change is not the volume of collection itself but the scope of items that require human review.
Although about 10,000 posts are collected daily, the number of posts that need human review is narrowed down to an average of about 30 per day.
Personnel's tasks shift from fixing crawlers and searching for posts to reviewing risky posts and making judgments on violations.
Simultaneously, information such as channels where high-risk posts are concentrated, trends in specific expressions, repeatedly detected sellers, and the proportion of risk by item become data that serves as the basis for managing priorities.
[Insert Image 1: Overview of Online Monitoring Dashboard]
[Insert Image 2: Distribution of AI Risk Levels and Proportion of High-Risk Posts by Channel]
Implications for institutions and companies from this case
| Target | Limitations of the conventional approach | Application Effects |
|---|---|---|
| Public agencies, local governments | Manual search of channels and posts by personnel | Management focusing on high-risk posts and repeated entities |
| Associations, review agencies | Focus on report submission and post-review | Member self-checks, analysis of industry-specific risk trends |
| Manufacturers, brand headquarters | Inspection focused on official advertisements | Inspection extended to distributors, sellers, and influencer content |
| Distributors, platforms | Difficulties in comprehensive inspection due to numerous products and sellers | Prioritizing inspection of risky expressions and sellers |
| Importers, distributors | Difficulty in identifying overseas sellers and unofficial distribution channels | Detection of unofficial distribution and unauthorized sales possibilities |
Public agencies and associations can continuously collect posts related to managed industries to focus on managing high-risk expressions, repeated exposure entities, and sudden increases in items or keywords. Associations can expand this to member self-checks, advertising review support, and preventive educational materials.
For companies, the scope of items to verify is not limited to ads created directly by the company. Instead of responding to regulatory inspections or consumer complaints after they occur, the focus is on establishing a pre-compliance system to verify the company and distribution network before issues escalate.
This is how it applies by industry:
- Food, health functional food: Checking for implied disease treatment, unauthorized functional claims, and misleading advertisements for general food products
- Pharmaceuticals, biotech, OTC products: Detecting unauthorized sales, parallel imports, abnormal overseas distribution, and exaggerated efficacy claims
- Cosmetics: Confirming unauthorized medical expressions, implied therapeutic effects, and unauthorized content from sales points, influencers
- Medical devices: Detecting advertisements for unauthorized use purposes and efficacy and unofficial sales channels
- Brand headquarters, distributors: Managing unauthorized expressions from agents, resellers, and confirming risky sellers at the entity level
Why keyword searches alone are not sufficient
Online sellers do not always use the same expressions. They may use circumventing phrases instead of direct disease names or include text within images. The meaning of the same word can vary depending on the item and context.
Therefore, keywords serve as the starting point for defining the collection scope, and actual risk analysis considers what items are being sold, what efficacy is implied, whether it encourages purchases and includes sales links, and whether the seller appears repeatedly.
When new slang or circumventing expressions emerge, the keywords and analysis criteria are continuously refined during operations.
Frequently Asked Questions
Why do institutions or companies transition from in-house crawling to professional services?
The primary reason is the maintenance burden. Whenever there are site structure changes, IP blocking, or captcha responses, the collection logic needs to be adjusted, and as the collection targets increase, personnel end up spending more time on crawler maintenance than on their core inspection and management tasks. Professional services handle collection and maintenance, allowing institutions and companies to focus on review and response.
Does automation eliminate the need for humans to inspect posts?
No. The role of automation is not to replace the final judgment but to reduce the review targets and establish priorities. In this case, AI analyzed about 310,000 posts and narrowed them down to about 1,000, leaving the final decision on violations to personnel.
Can the system detect sellers using slang or modified expressions?
It doesn't rely solely on keyword matching but also analyzes item categories, sentence expressions, sales situations, and context. However, since new slang continues to emerge, continuous refinement of keywords and analysis criteria is necessary during operations.
Can other institutions, associations, or companies utilize the same system?
Yes. By setting management criteria for target items, channels, and risk levels according to their tasks, institutions can use it for industry-specific surveys and member self-checks, while companies can employ it as a pre-inspection system focusing on their brands, product names, and sellers. The management scope extends beyond official ads to include sales points, resellers, open markets, and influencer content.
If the Ministry of Food and Drug Safety views data, institutions and companies should manage data as well
The case of the Ministry of Food and Drug Safety demonstrates a shift in the standards of online monitoring.
Regulatory agencies no longer limit themselves to manually searching for some posts. They continuously collect data from various channels and prioritize high-risk posts and repeated exposure entities for review, advancing their management systems.
Online compliance monitoring is not just a tool for post-inspection responses. It provides efficient online market management systems for public agencies and associations and a system for companies to detect regulatory risks in advance and manage their distribution networks.
If the Ministry of Food and Drug Safety is collecting and analyzing about 310,000 online posts monthly, our institutions and companies now need to manage online data as well.
Assessing Blind Spots in Online Management for Our Institutions and Companies
If you provide the managed product categories and online channels, we can review the following:
- Available online channels for collection and monitored brands, items, and keywords
- Risk analysis criteria and identifiable range of seller entities
- Dashboard configuration for institutions and companies
- Online management blind spots that require priority inspection
Check your online risk channels and applicable range.
Request a diagnosis of online management blind spots for our institutions and companies
If you are curious about the solution configuration, you can find it on the Online Compliance Monitoring Introduction page.
