Sift science skillfully using large data to prevent network fraud

For large data analysis, network security is a potential application field. By analyzing large data, the attacker's pattern can be screened to be more proactive in defending against network attacks, rather than being able to guard against known attack patterns, as is the case with traditional security defenses. And startups like Cylance, with big data technology, have done their financing.

The existing network payment and network transaction fraud prevention system is too complex, also can not effectively eliminate the network fraud, Siftscience such a start-up company began to try to use machine-learning based on large data analysis to prevent network fraud.

Recently, another company, Siftscience, uses large data analysis based on machine learning to provide services to protect against cyber fraud. Siftscience recently got the first round of investment in the 4 million dollar, led by venture capital fund Unionsquareventures. After counting the 1.5 million-dollar seed fund, Siftscience already has a total of 5.5 million dollars.

Siftscience's services are mainly for the Internet trading market, Electronic payment network and E-commerce sites, which is the most rampant network fraud. These sites can enjoy siftscience services simply by integrating a siftscience of JavaScript into a Web page.

Siftscience's co-founder, Brandonballinger, had worked for Google for four years before, and the main job was to guard against a large number of fraudulent ads. 5 of Siftscience's engineers are from Google, two are from the search department, and three have worked with Brandonballinger to guard against fraudulent advertising.

"We realize that every site on the internet will have some ' bad ' users, that is, some users who commit fraud," he said. The purpose of our creation of siftscience is to establish a fraud monitoring system. "Brandonballinger said.

Brandonballinger in June 2011 with his college roommate, Jasontan, who founded Siftscience. Initially, they financed the 2011 Summer project of Combinator. When they communicate with potential customers, the customer's initial response is that there has been a lot of fraud-proof systems already in place, and it seems that the problem with fraud has been solved.

"However, when we really have in-depth communication with our clients, we find that the problem is far from resolved," he said. Many websites buy fraud-proof systems, and almost no one really uses them. "Brandonballinger said. He pointed out that the existing fraud prevention system is still too complex:

They're not as easy to use as Googleanalytics or Mixpanel. Moreover, in order to use the fraud prevention system, you need to take a long sales process, need to have installation costs, need to have a minimum fee and so on. And the API is too complex. The existing fraud-prevention system uses soap, which often takes several months to integrate into the existing system. and Siftscience provides rest for this.

A big problem with existing systems is that they are using fixed rules. For example, they filter over a certain amount of trading, or a deal from Nigeria. However, the network fraud is not according to the fixed rules of the card, they change, you can easily through the change of behavior to cheat the fraud prevention system.

Siftscience uses machine learning algorithms to deal with cyber fraudsters. In Siftscience's database, there are more than 1 million patterns of network fraud, and it is being added through machine learning algorithms. For example, some URL browsing order, from the Tor node IP address, from late night transaction information, etc., may be added to the network fraud behavior patterns for analysis.

For each user, the site using siftscience can obtain the user's fraud-proof score through the API. Site can also be based on the machine learning model feedback, so that the fraud-proof model more suitable for the needs of the site.

This system can be used to detect fraudulent behavior in network transactions, or to help sites find spammers created by network fraudsters. According to Siftscience's statistics, their system is able to identify more than 90% of the network fraud behavior of the client site.

Siftscience's products have previously been tested by 20 customers, including Airbnb,uber and Listia. In addition, there are some top E-commerce sites and network payment platform. The pricing of Siftscience products is based on the number of user ratings that the site expects each month. Users below 5000 per month for free, more than 5000 users per month for 10 cents a user.

(Responsible editor: Fumingli)