BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pyconde-pydata-2026//speaker//8BCJ9W
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-pyconde-pydata-2026-DVCKHF@pretalx.com
DTSTART;TZID=CET:20260414T122500
DTEND;TZID=CET:20260414T131000
DESCRIPTION:Wolt’s Universal Venue Ranker (UVR) is a large-scale\, sequen
 ce-aware ranking model for personalized restaurant recommendations\, deplo
 yed across more than 30 countries. UVR replaces three previously independe
 nt models—Neural Collaborative Filtering\, a second-pass ranker\, and a 
 first-time-user model—by combining a transformer with a gradient-boosted
  decision tree for ranking.\n\nThe model follows a two-stage design. In th
 e first stage\, an encoder-style transformer learns a personalized user st
 ate representation from historical restaurant purchase sequences enriched 
 with spatiotemporal signals such as time and location. In the second stage
 \, a CatBoostRanker uses the transformer output as an input feature alongs
 ide additional user-\, venue-\, user–venue-\, and delivery-specific feat
 ures to score and rank candidate venues.\n\nIn this talk\, we present the 
 model and service architecture\, the training and evaluation setup\, and b
 oth offline and online results from a multi-country online A/B test\, demo
 nstrating significant improvements in global conversion rate and new venue
  trial rate. We also share practical lessons from deploying and operating 
 a multi-stage ranking model under strict latency constraints at global sca
 le.
DTSTAMP:20260412T141854Z
LOCATION:Palladium [2nd Floor]
SUMMARY:Personalized Restaurant Recommendations at Scale combining Transfor
 mer with Gradient-Boosted Ranking - Marcel Kurovski\, Steffen Klempau
URL:https://pretalx.com/pyconde-pydata-2026/talk/DVCKHF/
END:VEVENT
END:VCALENDAR