Deep Learning in eCommerce Text Search at eBay
In recent years, deep learning models became state-of-the-art in areas like speech recognition, computer vision, and more recently natural language processing. These models learn tasks through learning data representation rather than relying on heavy-handed feature engineering. eBay, with more than 1 billion listing items from 150 million sellers across the globe, stands to benefit immensely from such land-shifting scientific and technological advances. Search at eBay is one example. I will talk about how we apply deep learning to two search tasks. First, how we use query categorization to narrow searches to relevant subsets of item listings, greatly boosting search relevance. Second, how we use click prediction to improve search ranking.