Into the public comments mysterious team

For many consumers, objective and independent consumer review information can help make more accurate decisions. For third-party review platform, how to ensure that the evaluation of information independence, objective fact, a great challenge.

This reporter has learned that inside the public comment, there is a rarely known to the outside world mysterious force - integrity team, they are the world's police review, completely independent of the business team responsible for monitoring and handling all kinds of speculation, false reviews .

The face of counterfeiting practices and techniques of third-party speculation agencies emerging, how the unit is fighting with it, to ensure that the contents of the review of the independent, objective? With this curiosity, the reporter went into the public comment integrity team.

Technical algorithm with manual review to ensure accurate filtering

Like the police and the thief in the real world, it is not always peaceful in the world of integrity. There are always various kinds of cheating and anti-cheating fights. The team of honesty and trust is like checking the police in the world and maintaining a view of the world's independent and objective order. .

In the integrity team, the reporter saw the integrity of the member M Jun, she just pulled out from the screen full of data to interview, "review and handling all kinds of fake reviews and hype information is one of our very important tasks, every day from the number of Millions of reviews in the screening of these false information, and data analysis is one of the methods. " M Jun said that "at least need to read hundreds of data reports every day", in addition, her daily work needs to be done include browsing the system's feedback, handling customer complaints, focus on station forums, etc., and these are just The way to find the problem.

When asked whether it is necessary to manually identify one by one, M Jun laughed: "Of course not possible, the public commentaries have a sound system of integrity, mainly through technical algorithms to filter false reviews, manual review is only an assistant, if there is abnormal data , The system will automatically report to the police. "Currently 90% of the fake information system will automatically handle the remaining 10% M Jun and his friends need to conduct a manual review. "The screening of information is a meticulous and arduous task, and sometimes in order to find out the possible existence of false information, we also set ourselves the role of self-supposition from the false information publisher to predict." M also stressed The greatest challenge of honesty system lies in that it is necessary to ensure that it is fully grasped, grasped in time, minimize leaks, and there must be no mistakes. All the work is to ensure the accuracy of the filtering.

Weekly new algorithm attack speculation comment

In M Jun view, the integrity of the work group is actually a major cycle, namely: the discovery of the problem - data analysis - refining features - design algorithms - algorithms - validation on-line optimization system - multi-collection feedback found problems. The work of all around these are constantly breaking down and moving forward.

If we compare the first two steps to a police investigation of a thief, the next most crucial task is to seize cheating and carry out security upgrades.

After the investigation, M Jun and team members need to analyze from multiple dimensions to confirm the emerging issues whether there is a speculation comment feature, if any, the different characteristics of the problem will be extracted, the algorithm design, verification feasible Immediately after the line, which is the top priority of the integrity team.

"Based on the previous cheating and anti-cheat work, we now have about 100 algorithms," M-Jun said third-party speculation agencies and criminals follow up quickly, if found before the fraud rules will pass try to change other Way, "Every week there are still new algorithms to follow up speculation."

It is understood that the public comment has about 100 kinds of algorithms to filter fake reviews. In the meantime, she also conducted basic research on anti-cheat algorithms based on Bing Liu and its research team, rotating chairman of the Data Mining Association of the United States. She also drew on advanced algorithms for credit card fraud detection in the United States, technically walking on companies such as Yelp and Taobao front.