This document discusses Amazon Forecast and Amazon Personalize for improving personalization, recommendations, and forecasting using machine learning. It provides examples of how deep learning techniques can increase customer engagement by 15-29% and click through rates by 7-20% for recommendations. Amazon Personalize and Amazon Forecast use deep learning models to automatically generate personalized recommendations and accurate forecasts that consider multiple time series and additional metadata. They provide simple interfaces and pre-defined schemas for common use cases.
3. Common applications & use cases
Personalized
recommendations
Search
reranking
Notifications and
emailsRelated Items
4. Personalizing user experience is proven to increase discoverability,
engagement, user satisfaction, and revenue
30% of page views on
Amazon are from
recommendations
… However, most customers find personalization hard to
get right
5. Effective personalization requires solving multiple hard problems
Reacting to user interactions in real time
Avoiding mostly showing popular items
Handling cold start (insufficient data about
new users/items)
Scale
6. Deep learning techniques have a direct impact on the bottom line
SimilarityPopularity
Neural
network
Matrix
factorization
+15.4%
Engagement
Recurrent
Neural Net +
Bandit
Rule-based card
ranker
Bayesian
network model
+7.4%
Engagement+29%
Click Through
+20%
Click Through
7. Deep Learning delivers state of the art performance
0.954
0.928 0.925 0.922
0.91
0.856
Rolling
Average
T-SVD [2009] PMF [2008] RRN [2017] DeepRec
[2017]
HRNN
Ratings RMSE on Netflix
98 MM interactions, 500k users, 18k items
Rolling Average T-SVD [2009] PMF [2008]
RRN [2017] DeepRec [2017] HRNN
0.933
0.916
0.871
0.857
0.846
Rolling
Average
FM [2012] I-AutoRec
[2015]
RNN HRNN
Ratings RMSE on MovieLens
20 MM interactions, 173k users, 131k items
Rolling Average FM [2012]
I-AutoRec [2015] RNN
10. Feeding data to Amazon Personalize
Historical user
activity
User
attributes
Item
catalog
Real-time data
Mobile
SDKs
(coming soon)
JavaScript SDK
Amazon S3
bucket
Server-Side SDKs
Offline data
11. Train custom models once you ingest data
Use AutoML or pick a
predefined algorithm
recipes AutoML
Hyper
Parameter Optimization
19. Traditional time-series models
• Independent forecasts
• Strong structural assumptions
• De-facto industry standard
• Well-understood, > 50 yrs. research
• Data must match the structural
assumptions
• Cannot identify patterns
across time series
21. Traditional methods struggle with real-world forecasting
Can’t handle
time-series with
no history
Only process a
single time-series
at a time
Don’t consider
additional inputs:
related time-series,
metadata
Only predict a single
value: how trustworthy
is it?
25. … but using additional inputs helps to figure them out
26. Using additional inputs
• Additional inputs can
• Explain historical data
• Drive forecast behavior
• Examples from retail
• Price information
• Information about promotions
• Out-of-stock information
• Web page views
• Known future events
• Categorical inputs can be used to
identify group-level patterns
Fashion
Women’s
Clothing
Shoes
Watches
Men’s
Clothing
Shoes
Watches
Girls'
Clothing
Shoes
Watches
Boys'
Clothing
Shoes
Watches
30. Probabilistic forecasts
• Quantification of uncertainty
• Support optimal decision making
• Make “wrong” forecasts useful
• Forecasts can be obtained for
different quantiles of the predictive
distribution
p10: 10% of predictions with be lower
p50: the mean value
p90: 90% of predictions with be lower
p10-p90 interval: 80% of possible predictions.
31. Deep learning time-series models
• Global models: identify patterns using
all available time series
• Group-dependent seasonality and lifecycle
• Behavior in response to extra inputs
• Weak structural assumptions
• Can be significantly more accurate than
traditional methods
• Can easily incorporate and learn from
rich metadata
• Support cold-start forecasts for new
items
33. Amazon Forecast
Improveforecastingaccuracyby up to 50% at 1/10th thecost
K E Y F E AT U R E S
Consider multiple
time-series
at once
Automatic
machine
learning
Visualize forecasts &
import results into
business apps
Evaluate model
accuracy
Schedule forecasts
and model
retraining