When finance and technology are extremely hot, what can be done to control the wind under payment scenarios? (on)

This article is organized in part from the titanium media related sharing event.

“In the present day, the black-market fraudsters we face are characterized by vagrancy, specialization, and gangs. The industry needs to build a smart network for risk control. Through joint defense and joint control, fraudsters can no longer do it.” The founder and director of product anti-fraud and basic risk control Zhu Wei said.

With the rapid development of Internet technology, the beginning of low-level technologies, such as blockchain, has begun to be applied beyond bitcoin, such as cross-border financial payments. On the basis of pursuing efficiency, trust and risk control are topics that cannot be bypassed. Nowadays, risk control in the payment process includes big data and machine learning, computer vision and biometrics.

Using big data and machine learning

Zhu Wei said that the concept of risk control for big data focuses more on real-time risk analysis in the cloud, and it has found clues in the correlation analysis of user behavior data to prevent further fraud. The charm of cloud big data risk control is that even if the user side is already in an insecure state, for example, the user has leaked due to Trojan phishing or because a website is being dragged and the account password has been leaked, we can still judge through the data association analysis in the cloud. The account is abnormal and responds immediately.

The application of big data risk control in the payment industry, no matter which scenario, can use the data, based on the machine learning decision model to output the final fraud scores and recommendations in real time, and the customer should perform operations based on the risk decision results of the program.

Scenario 1: Registration scenario

For accounts with an account system, the registration scenario is mainly subject to the risk of fraudulent registration of spam. For example, fraudsters want to use small crazy registration to defraud merchants of promotional benefits.

Behind the various services, the fraud rules model is also more complicated. Zhu Wei said that in the registered security protection service, we can analyze the user registration information, user registration environment and user registration behavior, and then use machine learning. The decision model draws a scoring result. “In response to such frauds, we analyze the user registration behavior more abnormally in the cloud. For example, if the current registration request source IP address is the proxy, the registration initiated on the same device. Is the behavior too frequent?"

Scenario 2: Login scenario

For accounts with an account system, there are major risks of account theft and database crash in the login scenario. Then, in the login scenario, we can analyze the user's login behavior, login environment, and user habits. For example, "We used the rules model to calculate in real time that the interval between two logon times for a user is less than 2 seconds, but the offset from the login IP resolution location is more than 10 kilometers. If so, what do you think of this login behavior?"

Of course, our premise is to eliminate the special case of carrier address assignment drift through some IP technologies. In practical terms, this kind of mobile behavior has exceeded the speed range of the manned tools. The only possibility is to log in to the IP proxy and attempt to hide the login source.

Scenario 3: Payment scenario

In the payment scenario, the main risks faced by the platform are the fraudulent card payment and the anti-money laundering anti-cash monitoring required by the regulatory authorities. Then, in this scenario, it can be analyzed from the user's payment behavior, payment environment, and other dimensions. For example, a silent payment account suddenly receives a small payment, and several times later, the payment operation is equalized. Such behaviors are often highly risky. The fraudsters make initial activations after successful piracy, and proceed with bulk transfer after successful small-scale attempts.

Give another example: money laundering behavior is abnormal. Through the analysis of the inflow and outflow of account funds within one week or one month, if the flow of funds is concentrated in some accounts, and the active IP and equipment of these accounts are the same or similar, then the risk anomaly is very high.

Scenario 4: Credit payment scenario

In the credit payment scenario, in addition to the fraud risk, the credit risk of financial account holders is more concerned. In different stages, the focus of risk control is also different:

First, in the pre-grant letter stage , we can help the platform to make risk decision analysis for borrowers through a set of rules models. Through the analysis of the bad history of the borrower, if the borrower once had a dispute in the court such as dishonesty or enforcement, or the credit platform that has cooperated with the shield had overdue or lost performance, the risk of such user defaulting again. Is higher than normal users.

In addition, the use of data in the cloud to analyze whether users have recently had records of long applications and long liabilities on more platforms can also be used as risk control. The long application data can predict the user's desire for funds, and the liability data can predict the user's economic pressure. We can further make predictions of default risk through specific calculation data of long application and liabilities.

Secondly, in the stage of loan advancement , the borrower needs to be continuously tracked and managed after lending, that is, after the borrower has continuously monitored the loan, when the borrower experiences credit deterioration or change, the monitoring model will be the platform for the first time. Make risk warnings. For the platform, it is necessary to adjust the collection strategy according to the risk pre-warning situation.

Computer Vision and Face Recognition

Yang Fan, co-founder of Shang Tang Technology, stated that financial behavior is actually talking about the exchange of values ​​at different times and in different locations. It is actually a process of exchange of valuables between people.

There are several important points: First, you must first assess the value of the object and determine it. Second, although the thing you are exchanging is the object, the person actually exchanging is the person, that is, between the person and the object. Relationship binding. Only before and after the exchange of the two parties and the exchange of people, there is a correct binding between people and things. This financial behavior can only be completed in a complete and correct manner.

Is this binding always correct? actually not. First of all, the binding link between people and things, that is, the relationship between people and things, often becomes a very risky point in the whole financial behavior. For example, credit card fraud, one room and two sales, etc. It is the result that the binding relationship between people and things is stolen or falsified by others.

With the financialization of the Internet, the convenience and security of the binding of people and things have actually put forward higher requirements. In addition, the two indicators of convenience and safety are negatively related to a certain extent. When security is high, convenience is often reduced. However, we hope that the two indicators as a whole can have a common and continuous improvement. In addition, internet-based finance has ushered in even greater challenges in this regard. For example, when I use a mobile App to perform some financial activities, many times the traditional method is to send a verification code to my mobile phone. It will say that the verification code should not be told. In fact, many apps installed on the mobile phone will automatically read these verifications. The code has the function of automatically reading short messages, which creates a security risk.

Therefore, the ability of biometric identification technology represented by face recognition to provide remote personal identification verification for the financial industry is to bind people and things more effectively, accurately and quickly. So, how does face recognition apply to remote identity verification?

Please pay attention:

When finance and technology are extremely hot, what can be done to control the wind under payment scenarios? (under)

Posted on