Case Study: Protecting E-commerce with Counterfeit Goods Detection

Purpose: The proliferation of online counterfeit luxury goods poses a significant threat to consumers, brands, and the economy, leading to substantial financial losses for legitimate businesses. To tackle this growing problem Trilateral Research, an ethical AI solutions company, aimed to develop an AI-powered detection system to identify counterfeit fashion goods listings online. The sale of these goods violates intellectual property laws and inflicts harm on consumers, brands, and the economy [1]. To protect these stakeholders and support the economy, the developed tool collects information on luxury fashion goods sold online and assesses their authenticity, providing a Red-Amber-Green (RAG) rating for each product. The intended end-users of this tool are any organisation committed to combating online counterfeit goods.

Approach: The team created an end-to-end Python pipeline to detect and assess the authenticity of designer handbags sold online, focusing on designer handbags sold on online marketplaces. In the face of limited real-world data, the team demonstrated adaptability by overcoming this hurdle through extensive synthetic data generation and strategic data augmentation techniques. This pipeline leverages web scraping with the Scrapy Python library and machine learning, employing Natural Language Processing (NLP) techniques (word embeddings, sentiment analysis) to extract and engineer features from product and seller data. Unsupervised clustering (HDBSCAN) and supervised machine learning (Random Forest Classifier) were used to detect potentially counterfeit listings and provide RAG ratings.

The team also developed an initial pipeline for image analysis using pre-trained models (Roboflow and ResNet50) and explored additional online marketplaces (Amazon, Vinted) and ML algorithms to enhance the system's capabilities.

Outcome: The team successfully developed an end-to-end pipeline, demonstrating the feasibility of using AI to identify

counterfeit luxury goods sold online. The pipeline effectively scrapes product listings, preprocesses and models the data, and implements a RAG rating system to label suspicious listings. The project successfully identified sellers offering potentially fake designer handbags on eBay, demonstrating the tool's commercial potential.

The tool achieved high accuracy in counterfeit detection using listing text alone (e.g. reaching 85% in initial tests), effectively distinguishing genuine from potentially counterfeit listings. Initial image analysis provided highly performant results, identifying counterfeit logos with over 99% accuracy.

The tool offers significant value to stakeholders, including:

●      Customers: Minimises the financial risks (estimated in the millions of pounds annually [2]) and potential safety concerns associated with counterfeit goods.

●      Designer brands: Protects brand integrity and minimises revenue loss (estimated at 1.4% of UK manufacturing sales in 2021).

●      Online marketplaces: Enhances operational efficiency and maintains a trustworthy platform, crucial for retaining customer trust and market share.

●      The UK government: Safeguards the economy and public trust by reducing the negative impact of counterfeit goods, including an estimated £776 million in lost government revenue in 2021. [3].

Next steps: S2DS provided foundational work that unlocks Trilateral’s potential for scaling this tool and using wider datasets. Future development could focus on expanding detection capabilities across more brands, product categories and online platforms.