Abstract: A portion of gross merchandise volume in e-commerce is driven by categories such as fashion, home, furniture, lifestyle and apparel. These categories differ from other categories in terms of need for discovery and the way people making purchase decisions. They require visual appeal and visual discovery where people want to navigate them through how similar they are in terms of visual appearance rather than any other attribute of the products.
Most of the search engines and e-commerce search engines provide only text based searches and it utilizes textual attributes of products. This does not provide a good search and recommendations experience for categories that are rich in terms of visual information.
A visual search system that utilizes images of the products to provide search and recommendation results to end user enables
- a better experience in categories where visual appeal is very important to user
- better search results and recommendations where attributes of the products are not rich
This talk will give an overview of our visual search system that we built for Hayneedle catalog for furniture category which utilizes Kafka, Nomad, FAISS and Tensorflow.
Bio: Bugra is tech lead at Jet.com where he works on search and recommender systems. Prior to Jet.com, he was leading recommender systems at Hinge. He received B.S from Bilkent University and M.Sc from New York University focusing signal processing and machine learning.
He has two open source Python packages.