What is machine learning in the fight against counterfeiting?
Posted: Sun Feb 02, 2025 3:57 am
It seems that it is easy to isolate the price from the ad, but it is not. It happens that one ad contains many prices: the cost of the product, delivery, the price with and without a discount, the price for two or three products, and there is also a wholesale price.
The second problem is currency. Dollars can be not only American, but also Australian. The dollar sign is used in Brazil to denote the Brazilian real. The algorithm must understand in which country the product is sold.
Image
It is easier for a machine learning algorithm to work if the seller saudi arabia number data has left a detailed description of the product in the ad. But, for example, social networks often only provide a photo of the product and its price.
However, the product category and brand can also be determined from the photo. Counterfeiters often copy product designs, but they can make mistakes, for example, in color shades. The algorithm learns to determine this. After a person confirms several matches as signs of counterfeiting, new criteria will be created, and the algorithm will make its own decision.
Description
The neural network analyzes keywords in product descriptions. For example, a well-known brand does not need to write the phrases “excellent quality” and “directly from the factory.” This is self-explanatory. If two of these phrases appear in one ad, the algorithm will mark such an ad as “problematic.”
Algorithm training and accuracy
The algorithm is given a manually tagged sample of ads to start working. Training begins. When the algorithm finds a potential intellectual property violation but doesn’t know what to do with it, it alerts the operator. Analysts notice the data and feed it back to retrain the system.
The second problem is currency. Dollars can be not only American, but also Australian. The dollar sign is used in Brazil to denote the Brazilian real. The algorithm must understand in which country the product is sold.
Image
It is easier for a machine learning algorithm to work if the seller saudi arabia number data has left a detailed description of the product in the ad. But, for example, social networks often only provide a photo of the product and its price.
However, the product category and brand can also be determined from the photo. Counterfeiters often copy product designs, but they can make mistakes, for example, in color shades. The algorithm learns to determine this. After a person confirms several matches as signs of counterfeiting, new criteria will be created, and the algorithm will make its own decision.
Description
The neural network analyzes keywords in product descriptions. For example, a well-known brand does not need to write the phrases “excellent quality” and “directly from the factory.” This is self-explanatory. If two of these phrases appear in one ad, the algorithm will mark such an ad as “problematic.”
Algorithm training and accuracy
The algorithm is given a manually tagged sample of ads to start working. Training begins. When the algorithm finds a potential intellectual property violation but doesn’t know what to do with it, it alerts the operator. Analysts notice the data and feed it back to retrain the system.