10. Overview: Retrieval
Venue
Index
Check-in
Candidate Venues History
Venue Social Personal
attributes Interactions Interactions
11. Overview: Ranking / Reasons
Venue Social Personal
attributes Interactions Interactions
Personal
Global scores Social Scores Scores
Top K venues
Global Social Personal
reasons reasons reasons
12. Global attributes
• Time decayed popularity
• Time of day / day of week filtering
• Total interactions (# tips, todos etc)
14. Social / Personal scores
• # Friends that have checkins
• Tips left by friends, etc
• Prior visits by user
• Similar places user has been in the past
15. Venue Similarity
vi vj
Venues
for all i,j
sim(vi, vj)
Users
Incredibly Sparse
Matrix
16. Venue Similarity w/ MapReduce
key
user visited venues map
vi, vj score emit “all” pairs of visited venues
for each user
vi, vj score
...
key
vi, vj score score ... score reduce
Sum up each user’s score contribution
final score to this pair of venues
22. Interesting phrases
• Hack day project on tip text
• Build a n-gram model, find “improbable”
phrases relative to background corpus
• Inspired by Amazon’s statistically
improbable phrases (SIPs) feature
28. City level phrases
fried chicken, blue bottle, french toast
breakfast tacos, mexican martinis, fried
avocado
deep dish pizza, celery salt, hot dog
clam chowder, lobster roll, red sox
bread pudding, cafe au lait, bloody mary