Learning from Multi-User Activity Trails for B2B Ad Targeting - - PowerPoint PPT Presentation
Learning from Multi-User Activity Trails for B2B Ad Targeting Shaunak Mishra*, Jelena Gligorijevic, and Narayan Bhamidipati Yahoo Research, Verizon Media 1 B2B products Business-to-business (B2B) versus business-to-consumer (B2C) Same
1
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
2
❖ Business-to-business (B2B) versus business-to-consumer (B2C) ❖ Same advertiser can have B2B and B2C products
WiFi service (advertiser)
small-medium enterprise (B2B) home internet (B2C)
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
3
unaware aware consider intent buy
c
s u m e r j
r n e y
p u r c h a s e f u n n e l s t a g e s
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
4
unaware aware consider intent buy
c
s u m e r j
r n e y
p u r c h a s e f u n n e l s t a g e s visiting websites, forums, tech blogs
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
5
unaware aware consider intent buy
c
s u m e r j
r n e y
p u r c h a s e f u n n e l s t a g e s visiting websites, forums, tech blogs
More details on B2C in poster # 76 KDD Applied Data Science
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
6
unaware aware consider intent buy
analyst, IT manager
influence
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
7
❖ Conversion prediction is crucial for ad targeting ❖ Will conversion prediction suffer from multi-user activities for B2B purchase?
conversion prediction model
user 1 user 2 user 3
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
8
❖ Conversion prediction is crucial for ad targeting ❖ Will conversion prediction suffer from multi-user activities for B2B purchase?
conversion prediction model
user 1 user 2 user 3
Assumptions ❖ No declared profiles! ❖ Online activity trails available
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
9
❖ Conversion prediction is crucial for ad targeting ❖ Will conversion prediction suffer from multi-user activities for B2B purchase?
conversion prediction model
user 1 user 2 user 3
user 1 trail: act1, act2, search WiFi offers, act3, act4, purchase user 2 trail: act2, act4, search WiFi offers, act5, act6 user 3 trail: act1, act3, act5, purchase
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
10
❖ Conversion prediction is crucial for ad targeting ❖ Will conversion prediction suffer from multi-user activities for B2B purchase?
conversion prediction model
user 1 user 2 user 3
user 1 trail: act1, act2, search WiFi offers, act3, act4, purchase user 2 trail: act2, act4, search WiFi offers, act5, act6 user 3 trail: act1, act3, act5, purchase
Model can relate “search WiFi offers” to purchase Model cannot relate “search WiFi offers” to purchase
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
11
❖ Conversion prediction is crucial for ad targeting ❖ Will conversion prediction suffer from multi-user activities for B2B purchase?
conversion prediction model
user 1 user 2 user 3
user 1 trail: act1, act2, search WiFi offers, act3, act4, purchase user 2 trail: act2, act4, search WiFi offers, act5, act6 user 3 trail: act1, act3, act5, purchase
Model can relate “search WiFi offers” to purchase Intuition: augmenting user trails may help!
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
12
user 2 trail: act2, act4, search WiFi offers, act5, act6 user 3 trail: act1, act3, act5, … purchase Consider users in the same cluster/household
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
13
user 2 trail: act2, act4, search WiFi offers, act5, act6 user 3 trail: act1, act3, act5, … purchase relevant act (advertiser specific)
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
14
user 2 trail: act2, act4, search WiFi offers, act5, act6 user 3 trail: act1, act3, act5, … purchase relevant act relevant user
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
15
user 2 trail: act2, act4, search WiFi offers, act5, act6 user 3 trail: act1, act3, act5, … purchase relevant act relevant user augmented user 3 trail: act1, act3, act5, [ act2, act4, search WiFi offers, act5, act6 ], … purchase Model can relate “search WiFi offers” to purchase!
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
16
users in an org (commercial IP, household ID) relevant acts seed list
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
17
users in an org (commercial IP, household ID) relevant acts seed list Identify relevant users
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
18
users in an org (commercial IP, household ID) relevant acts seed list identify relevant users augment user trails in cluster by relevant users’ trails + use in conv. model
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
19
users in an org (commercial IP, household ID) relevant acts seed list identify relevant users augment user trails in cluster by relevant users’ trails + use in conv. model
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
20
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
21
user i trail: act1, act3, act5, … act10, act1, act2 activity2vec = word2vec style training with activity trails session = sentence act = word user trail = document
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
22
user i trail: act1, act3, act5, … act10, act1, act2 activity2vec = word2vec style training with activity trails session = sentence act = word user trail = document
neighbor based iterative seed expansion
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
23
initial seed list vs. no augmentation noise in expanded seed list
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
24
initial seed list vs. no augmentation noise in expanded seed list
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
25
Mishra, S., Gligorijevic, J., Bhamidipati, N. “Learning from Multi-User Activity Trails for B2B Ad Targeting”
26