Context-Sensitive Autocomplete Suggestions Using LSTM and Pair-Wise Learning at Target

Presented at virtual Activate 2020. Autocomplete is a predominant feature in e-commerce search. By being relevant, autocomplete should help users find the query they intended to type quickly and with minimal keystrokes. This talk is about how Target achieves autocomplete by considering the user’s context as a signal for re-ranking query suggestions. The context is based on a diverse sequence of events performed by a user on the website. It is generated using an LSTM model & fed into the Autocomplete service which re-ranks suggestions accordingly. This entire system is designed to work at a high scale in real-time, with low latencies.

Speakers:
Dileep Patchigolla, Lead AI Scientist, Target
Manohar Sripada,  Senior Engineering Manager, Target

Intended Audience:
Anyone interested in Machine Learning (ML) techniques for search relevancy (e-commerce or other, and a basic ML background would be helpful) and anyone looking to improve their search autocomplete experience.

Attendee Takeaway:
Gain insights into how the Autocomplete experience is powered at Target using state-of-the-art deep learning algorithms and how it is served in real-time at scale.