Challenge

Targeted Detection of Online Shopping Scams

The first step toward protecting consumers from fraudulent shopping websites is identifying and classifying sites among a vast array of websites and website categories. However, after rigorous testing, Ms. Bitaab found that widely used mitigation methods, such as Google Safe Browsing and phishing detection systems, did not accurately and precisely detect online shopping scams.

The researcher realized the need to create a new solution that collects domains daily, classifying them, filtering out unrelated ones, and analyzing the content of websites categorized as shopping.

Solution

Reliable and Fast Website Categorization API

The researcher used Website Categorization API to aid in her daily web classification process, allowing her to design efficient data collection pipelines for the online shopping category.

Website Categorization API’s accuracy and fast inference speed enabled the researcher to classify domains and filter out those unrelated to shopping. The API’s confidence scores for each website’s categories further allowed the researcher to set desired thresholds for better accuracy, ensuring the system can adapt based on unique confidence requirements.

Results

More Efficient Data Pipelines and Effective Online Shopping Fraud Detection

Accurate Detection of Shopping Websites

Website Categorization API is instrumental in the project’s primary process, which involves collecting domains daily, classifying them, and picking out only shopping websites.

Efficient Data Pipelines

Website Categorization API enabled Ms. Bitaab to optimize the designed data pipelines and focus solely on shopping websites. That translates to a smaller amount of data that needs to be processed, making the system faster.

Improved Scam Detection Classifier Accuracy

The inclusion of non shopping websites in the data pipeline would have decreased the accuracy of the scam detection system since their features differ from those of shopping websites for which the system was designed. By filtering out unrelated sites, Website Categorization API was able to help train and enhance the system’s accuracy.