AWS Personalize-Admin-Manual

anonymous·2022년 9월 7일

CREATE

1 Create a domain dataset group

Register Data in S3 bucket
Required Data- User, Item, Interaction (3-MAIN-DATA)

2 Import Data

Get Data from S3 Bucket

3 Create recommender / Create Event Tracker

Create and test recommender (Takes some time for deep learning)
Create Solution
Create Campaign
(Warning - High Costs for having each of these solutions)
Create Event tracker -> Connect with kinesis stream-putevents API for real time event
Apply filters for recommendations filter
SCHEMA


EXCLUDE/INCLUDE ItemID/UserID WHERE dataset type.property IN/NOT IN (value/parameter)

4 Get recommendation

5 Connect with Other AWS Infrastructure with Lambda functions

Watch Event Put with CloudWatch/CloudTrail tracker
https://docs.aws.amazon.com/personalize/latest/dg/personalize-monitoring.html

6 Clean up All resources after use (High maintainenance cost)

READ

Recommendation Types

User-Interaction Personalization
	- Use item-user interaction (historical and real-time interaction) weight data 
	- find relevance 
Popularity Count
	- Use item-user interaction data 
	- find popularity 
Related Items 
	- Generete recommended items based on interaction data, item metadata 
User Segmentation 
	- Segments user based on item input data 
	Item-affinity 
		- User segment for specified item 
	Item-attribute-affinity 
		- User segment for each specified item attribute

Data Type (users, items, interactions)

USER ID, ITEM ID, EVENT TYPE, EVENT VALUE, TIMESTAMP

Filtering Recommendation Result

Filter value by type (genre)

User Feedback

Explicit 
	- System asks user rating for items 
Implicit 
	- System tracks user behavior, preference. (No direct user participation)
Hybrid
	- Utilize both explicit and implicit feedback