How to ensure the safety of the tip of the tongue by big data

Law enforcement personnel of Shushan District Market Supervision Bureau of Hefei City carry newly configured food safety rapid detection instruments, spot check all kinds of food in supermarkets and shopping malls under their jurisdiction, and produce results on the spot. Bright Picture/Vision China

Recently, the National Food Safety Demonstration City and Agricultural Product Quality Safety County were established in Chengdu, Sichuan Province. How to Eat Safer in the Internet Age? Experts at the meeting offered suggestions and suggestions, proposing to pay attention to the use of modern information technology means to innovate supervision methods and improve supervision efficiency.

As one of the most popular information technologies in recent years, how can big data technology improve the regulatory effectiveness of food safety? In Chengdu, which is creating a food safety demonstration city, we see a possibility.

Open up information islands

Unlike many food safety regulators, Wang Lishan doesn't need to go to farmers 'markets, supermarkets or restaurants. He sits at a table most of the time. His "inspection" object is a pile of data. "Data analysis can improve the efficiency of food safety regulation." he said.

Wang Lishan is director of the Chengdu City Food Safety Risk Monitoring Data Center, which is creating the core technical support system for Smart Food Safety. "Smart regulation, data first", that's their slogan.

Walking into the data center, the most striking thing is a black electronic screen, on which the data is constantly refreshed. At this moment, the data on the screen shows that the total number of test batches is 674136, the total number of test items is 8469618, the total number of unqualified samples is 15619, the number of unqualified samples to be processed is 4, the number of samples being processed is 3...

These figures are not exhaustive. Wang Lishan and his colleagues face data not only from various food testing institutions, but also from local agriculture, quality inspection, health, industry and commerce departments. The data includes food detection information, food safety public opinion information, enterprise industrial and commercial registration and credit data, foodborne disease case data and environmental data.

"Chengdu has nearly 200,000 food production and operation entities. From farmland to dining table, every link is scattered with huge amounts of data, and fragmentation is serious." Wang Lishan said that what they need to do is to integrate these data, establish basic databases, open up isolated information islands, and realize data exchange and resource sharing.

At present, the database has accumulated 70,000 pieces of shared data of government departments, 8.6 million pieces of inspection and detection data, accessed 300,000 pieces of basic data of manufacturers and 70 million pieces of global commodity data, synchronously monitored public opinion data of multiple online media, and maintained an average monthly growth rate of 200,000 pieces.

"But big data is not equal to a large amount of data, but to produce a value of 1 plus 1 far greater than 2." Wang Lishan said.

They did it. From January to August this year, through big data mining and analysis, the discovery rate of food safety supervision sampling problems in Chengdu City increased from 2.36% to 10.71%, and the supervision efficiency was greatly improved.

So, how exactly did they do it?

Achieve targeted strikes

At four o 'clock in the morning, Chengdu Agricultural Products Center Wholesale Market began to be busy. Trucks loaded with all kinds of vegetables drove into the market and split into two trains. One train has magnetic cards, the other doesn't. The magnetic card contains the registration information and safety certificate of the incoming goods of the vegetable merchant. For trucks without magnetic cards, the goods will be sent to the fast screening testing laboratory, and the inspection data will be automatically uploaded to the food safety risk monitoring data center of Chengdu City dozens of kilometers away.

Here, the results can be obtained in 24 minutes by quick inspection, and more than 600 batches can be inspected every day. After the detection information is uploaded to the data center, the effect is quickly reflected in the next link.

Those trucks equipped with magnetic cards, although with safety certificates, still need to do a certain proportion of random inspection according to regulations. So encounter a difficult problem, which vegetables should be sampled most likely to find problems?

Wang Lishan said that through data analysis, they found that unqualified food is highly unevenly distributed geographically and seasonally. In different seasons and different climatic conditions, foods prone to safety problems are different. Moreover, enterprises with single shareholder structure and no corporate shareholders and low paid-in capital are more likely to be detected unqualified food. This inspired them to integrate various types of inspection data, business registration information and credit data to build a machine learning model. This work was done jointly with the team at the University of Electronic Science and Technology.

"Through certain algorithms, the machine learning model can quantitatively assess the possibility of unqualified inspection of the product to be inspected, so as to realize targeted sampling inspection for high-risk food." said Zhou Tao, chief expert of the data center and professor of the University of Electronic Science and Technology. The wholesale market information platform is connected with the data center, and the high-risk food information obtained from data analysis is directly transmitted to the wholesale market.

"We're going to receive data from them every day on high-risk products, and that data will make our safety oversight more effective." Wholesale market deputy general manager Zheng Keke told reporters.

The data centre also consolidates case data on all foodborne gastrointestinal diseases in the health sector. By monitoring seasonal and weather changes in the number of cases of infection by different species and analyzing 8.6 million test data, they obtained associations between food and detected species, thus establishing an automatic decision-making channel from related cases to targeted sampling, Wang said.

This kind of targeted sampling based on big data analysis finally enabled them to obtain the regulatory efficiency mentioned above and achieve more accurate attacks on unqualified food.

Early termination of rumors

Finding food safety problems more efficiently is just one of the magic of big data. Zhou Tao and Wang Lishan and other once confused is, for the network every day circulating a variety of food safety rumors and their panic to the public, can also timely monitoring, or even early termination?

They monitored hundreds of well-known websites, forums, postbars and other accumulated more than 3 billion pieces of public opinion data, including nearly 8 million pieces related to food safety. "After obtaining public opinion information, through background algorithms, public opinion events are extracted and automatically clustered according to key semantics, and we can judge the trend of public opinion transmission." Zhou Tao said.

Next, it is a more critical step: divide all the public opinion data mined into positive emotions and negative emotions, and analyze the intensity of positive and negative emotions. "Through this analysis, we can evaluate the risk level of a food safety public opinion, and whether the regulatory authorities should actively intervene to eliminate rumors and potential safety hazards through timely inspection or strict law enforcement." Zhou Tao introduced.

Not long ago, they first time monitoring found that Chengdu area network spread "cabbage dipped in formaldehyde to keep fresh". At that time, this public opinion has not spread widely, but they think the risk level is very high after analysis, timely notify the regulatory authorities to go to the market for large-scale spot check cabbage. "In the end, no formaldehyde cabbage was found. The authoritative inspection report was also publicized to the notary public to terminate the rumor ahead of time." Wang Lishan said, They also through data analysis timely found flour products aluminum, Aquatic products malachite green and other food safety hazards 30 items, Timely carry out targeted remediation, Eliminate safety risks.

In fact, grasp the risk from the beginning, this kind of "pass forward" practice is not unfamiliar to many regulatory departments. "However, most of the methods are relatively backward, the amount of data is small, and people rely on people to pay special attention to some websites and report regularly. Now that the data is getting bigger and bigger, it's not realistic to rely solely on people." Zhou Tao said.

Data is an important resource.

What makes Wang Lishan even more excited is that data analysis can also actively discover possible defects in current standards and optimize inspection standards.

Their database contains a large amount of detection information about water-injected meat. After analysis, it is found that the inspection data of pork moisture content are concentrated around 63% and between 73% and 77%, while the national standard is 77%. "Many merchants just inject pork near the standard line and pass the test. But in fact, pork with 76% water content is already seriously water-injected meat." Wang Lishan said.