Scenario Description: on Amazon, a well-known e-commerce platform, fake reviews have been inducing consumers to buy fake and inferior products. Although Amazon has some control over this, it is still unable to eliminate this phenomenon. Recently, it has been reported that Amazon’s sellers are buying a lot of fake comments from professional counterfeiting companies at a high price, and even forming a large gray market. How to identify and supervise these fake comments? What kind of role can AI play in it?
Key words: e-commerce shopping false comments natural language processing
All along, Amazon has been busy fighting false reviews.
- Amazon has implemented a lot of regulatory measures, but there are still many false comments
Recently, the daily mail, a foreign media, published a survey. On Amazon, there are a group of companies that sell fake comments. They manipulate the fake comments on the website behind them. Each fake comment is sold for 13 pounds (about 116 yuan).
Among the revelations, AMZ tigers, a German based company, specializes in such improper transactions. It has 3000 testers in the UK and nearly 60000 in Europe to provide fast false comments.
Amazon has repeatedly said it has implemented regulatory measures on the platform, and has spent 300 million pounds (2.726 billion yuan) over the past year to protect customers from abuse, fraud and other forms of misconduct. But in this report, there are still some loopholes in their work.
- The British consumer association has angrily labeled many Amazon products “don’t buy”
False comments are not only a problem that Amazon needs to face. In fact, there are a lot of fake / fraudulent comments in the whole e-commerce platform, which is almost the lingering haze in the online shopping environment.
False comments following the spread of online shopping
Behind the false comments, there is a huge market driven by interests and vicious competition.
In a survey of 2000 adults, more than 97% of buyers rely on the content of online reviews to make final purchase decisions.
False comments are often accompanied by swiping. At the end of last year, a survey by the new Beijing News mentioned that in 2018, Alibaba single monitored more than 2800 order swiping groups, including 2384 QQ groups, 290 empty package trading platforms and 237 order swiping trading platforms.
The threshold of single brush rating is very low, which can be seen everywhere on the Internet, and also controls huge traffic. For example, the large platform “handshake network” boasts 600000 “brush hands”, while “baby brush single network” claims nearly 10000 “brush hands” online every day. After the exposure, some platforms were shut down, others changed their faces and started the business again.
- The domestic e-commerce platform also has the blatant behavior of single brush review
CCTV has also exposed that a company selling substandard products for children’s products has sold 1231 orders in one year, involving more than 770000 yuan, forged nearly 40 times of false transaction records, and paid about 20000 yuan of Commission for such orders.
E-commerce, online “brush hands” and “brush bill” channels, interwoven together, constitute a huge black industry chain of brush bill. They have a set of complex and detailed operation process, resulting in a large number of false sales and false comments.
- Even bad reviews can be fake, CCTV reported
On January 1, 2019, the e-commerce law came into force, in which “bill swiping” was defined as an illegal act, stipulating that “e-commerce operators shall not carry out false publicity by means of fictitious transactions and user evaluation, so as to deceive and mislead consumers. However, due to the characteristics of low illegal cost and strong concealment, as well as the recommendation mechanism of the platform, the barbaric growth of this phenomenon was born.
In addition to the need for e-commerce platform to learn from more perfect management system, perhaps another feasible way is to make good use of the power of AI.
AI technology is becoming a weapon to fight against counterfeiting
Since people and e-commerce platforms attach importance to false comments, traditional data analysis methods have been used to detect fake / fraudulent comments. However, the early data analysis technology usually focuses on the extraction of quantitative and statistical data features.
- Repeated fake comments on Amazon
These methods can screen out some low-end forgeries. In order to carry out a more comprehensive and deep data analysis, the system must be equipped with a large number of background data and be able to perform the reasoning tasks involving the data.
As a result, some researchers turn to machine learning and artificial intelligence, and use more effective methods to fight against false comments.
- Fakespot is an AI based online detection tool, which supports Amazon, steam and Wal Mart’s review detection
There are some linguistic differences between real and false comments. For example, a newly registered user is full of praise words in the comments, so it is likely to be comments from the Navy. Using supervised / unsupervised learning methods to let AI learn to judge these differences is a key step for AI to fight against counterfeiting.
This kind of technology belongs to the category of natural language processing. Through the extraction and identification of popular e-commerce platform, false and unreliable consumer comments, through training let algorithm learn to identify and judge false comments, and finally learn to score the credibility of user comments.
One of the important tasks is to detect exceptions, and judge whether the comment belongs to a normal user or a malicious fake comment from the writing style, format and various available materials of the reviewer.
For example, find the spelling and grammar suspicious, combine the number of comments, purchase method, date mismatch and other signs of suspicious comment activity. Combined with the human analysis team, it analyzes from multiple data dimensions to determine the authenticity of the comments.
- Fakespot can detect the virtual high evaluation and give the real user’s score
In addition, there are some more detailed studies, which will make more accurate judgments on comments based on the reviewer’s comment density, emotional analysis and semantic detection.
Healthy online shopping environment, can AI give?
When we try to find more product information from user comments, those who have ulterior motives have already used them as chips for illegal profits.
AI can help us filter out the false comments from these platforms and create a relatively clean online shopping environment.
At the same time, AI will also be used to make false comments, and then the technology of detecting AI fraud will be born… This battle is gradually upgraded to the battle of AI attack and defense.
But it is certain that only by making good use of technology and using the right technology can we achieve the final victory.