VERD: Emergence of Product-Based Video E-Commerce Retrieval Dataset from User's Perspective

Area of Science:

• Computer Science
• Information Retrieval
• E-commerce Technology

Background:

• The e-commerce market is rapidly expanding, increasing customer demand for effective product search.
• Current object-centric retrieval methods using product images struggle with complex user environments and large data requirements.
• There is a need for advanced retrieval techniques that can handle the nuances of real-world e-commerce interactions.

Purpose of the Study:

• To introduce a novel dataset, the Video E-commerce Retrieval Dataset (VERD), for product search.
• To facilitate research in product-centered video-based retrieval.
• To address the limitations of existing image-based retrieval systems in e-commerce.

Main Methods:

• Development of the Video E-commerce Retrieval Dataset (VERD) utilizing user-perspective videos.
• Establishment of a benchmark for evaluating video-based product retrieval.
• Conducting additional experiments to validate the dataset and retrieval approach.

Main Results:

• The VERD dataset provides a valuable resource for studying product retrieval in complex scenarios.
• The benchmark demonstrates the effectiveness of user-perspective videos for e-commerce search.
• Experimental results highlight the need for dedicated research in video-based product retrieval.

Conclusions:

• The Video E-commerce Retrieval Dataset (VERD) is a significant contribution to the field of e-commerce search.
• VERD enables more robust and context-aware product retrieval compared to image-only methods.
• The dataset is publicly available for academic research to foster further innovation.