1. Do you know that feeling when an app
knows exactly what you want?
@renan_oliveira
Principal Data Scientist
2. AN UNIQUE MOMENT - YOU CAN'T FAST-FORWARD OR SKIP
Time and Quality
Decision process
Special occasion
We are not a streaming service.
Ordering food has an additional
difficulty being assertive because
fixing a mistake is not like skipping
a song: placing an order involves
more money and logistical operation.
3. IFOOD IN NUMBERS
20 million orders
270k drivers
100k restaurants
660 cities
https://www.uol.com.br/tilt/noticias/redacao/2019/08/29/haja-fome-ifood-recebe-mais-de-7-pedidos-a-cada-segundo-no-brasil.htm
4. FOOD IS A VERY PERSONAL CHOICE
Taste Profile
Speed
Brand
Price Affinity
Offer Affinity
etc…
Taste Profile
Dish
Cuisine
Offers
Rating
etc…
Match
5. CHALLENGES
Locality geographical constraint for model training
Speed if you are hungry you will want to eat soon
Serviceability production capacity and restaurant quality
Feedback implicit (engagement) vs explicit (ratings)
Growth cold start problem
7. COLLABORATIVE FILTERING - IMPLICIT FEEDBACK
5 2 2
? 3 4
5 3 ?
ASH
GOKU
SEIYA
Orders
User Journey
Built engagement
ALS is very good to reinforce
tastes but fails to identify new
tastes. Since our problem has
geographical constraints, our
data is more sparse than usual.
8. USER HISTORY + CONTENT BASED
BOUGHT
TOMATO PIZZA - 3
FREE DELIVERY - 2
TOMATO PIZZA - 2
FAST DELIVERY - 1
TRENDING - 2
HAMBURGUER - 4
COKE - 2
FAST DELIVERY - 1
TERMS AND
RELEVANCE
TOMATO PIZZA - 5
HAMBURGUER - 4
FAST DELIVERY - 4
COKE - 2
FREE DELIVERY - 2
FAST DELIVERY - 3
TRENDING - 2
HISTORY
TOMATO PIZZA
FREE DELIVERY
COKE
SALAD
JUICE
FAST DELIVERY
TOMATO PIZZA
HAMBURGUER
FAST DELIVERY
TERMS STORESRECSYS
9. “AKINATOR” - EXPLICIT FEEDBACK
1. Select cuisines
2. Select the expected price
3. Select time range
Kmeans is great for a cold start
scenario in which we don't know
much about the customer but we
have information about other
costumers and restaurants nearby.