CASE STUDIES
E-Commerce Product Categorization and Tagging
A Leading E-Commerce Marketplace
Labelbox
Objective
To streamline the categorization and tagging process for newly uploaded products, enhancing search accuracy, user experience, and inventory management.
Process:
- Data Annotation: Utilized Labelbox to categorize and tag products based on images and descriptions, ensuring precise annotations for both visual and textual data.
- Human-in-the-loop (HITL) Annotation: Human annotators reviewed AI outputs, addressing ambiguities in product descriptions to improve AI accuracy.
- Continuous Feedback: Established a feedback loop where annotator corrections were used to refine the client’s machine learning model, reducing human intervention over time.
Challenges:
- Ambiguous Descriptions: Multi-functional products (e.g., Bluetooth speaker lamps) required human review for accurate categorization.
- Multilingual Data: Human experts handled product descriptions in multiple languages to ensure accurate classification across regions.
- AI Model Bias: Initial AI outputs favored popular categories, which human annotators corrected to ensure balanced tagging.
Outcome
Achieved a 35% improvement in categorization accuracy within six months and reduced the human review process by 40% for straightforward cases, maintaining high precision in complex descriptions.
