Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

WebCenter PS3 Master Workshop

<Insert Picture Here>

Day 5: Activity Graph


WebCenter Product Management
What is the Activity Graph?

• Activity Graph is an extensible data model that captures people, their


relationships, their activities and the artifacts they impact
• An analysis engine that infers new, relevant connections to people and
artifacts.
• Engines analyze the graph to facilitate recommendations, intelligent search,
and expert ID
Contextual Recommendations on the Web

• Consumer sites calculate similarity scores between items by


analyzing user activities like purchasing
• End users benefit by seeing contextual recommendations on
product details pages
Data Model

Profile & Connections Activities Content

• Tom works in Product • Colleen liked a • Spaces


Management - profile document
• Discussions
• Tom knows Colleen - • Colleen checked in a
connection document
• Documents

• Colleen works for • Colleen searched for


Pankaj - derived “project status” and
clicked on a link
Analysis Engine

• Public-domain
algorithms

• User-defined weighting
– “recipe”

• Yields new, relevant


relationships

• Extensible
What can it do?

• Ranking
– Who are the most influential people and what are the most
important documents and business objects
– More relevant search results
– Expert Identification
• Recommendations
– People who worked with this item also worked with these
others
– Here are the people you should know
– Here are the items people like you are working with
Recommendation Task Flows

• Similar Items – People who worked on this document also worked on these
documents, wiki pages, discussion topics, etc.
• Similar Spaces – People who worked in this Space also worked in these others
• Recommended Connections –
These people visit the same Spaces
and work on the same items as you

• Each Task Flow offers:


• Recommendation reasons
• Ability to remove unwanted
recommendations
• Customizable recommendation
recipes
Search

• Activity Graph calculates query-independent ranks and pushes


them into the SES index to improve relevance
• Query-independent ranks are combined with query-dependent
scores at search time to order results
• Top results are the items that match your query and which others
use and find useful
The Magic
Recommendations

• Similarity Calculations
– Items are modeled as vectors of people
– People are modeled as vectors of items
– A Similarity Calculation is just a vector space metric
– We use „Tanimoto‟ for WebCenter integration, but also support
Cosine and Pearson coefficient

• Recipes
– Our query API accepts weighted lists of Similarity Calculations
– Example: Recommended Connections Task Flow
• Mutual connections – 100
• Items edited in common – 50
• Items tagged in common – 10, etc.
Some Statistics Stuff
Well You Did Ask! 
Cosine Similarity & the Tanimoto Coefficient

• “Cosine Similarity” is a measure of similarity between two vectors of


n dimensions by finding the angle between them. This is often used to
compare documents in text mining (Activity Graphs).

A B
cos( )  
A B
• The “Tanimoto Coefficient” is an extension of the cosine similarity
that yields a statistic used for comparing the similarity and diversity of
sample sets (“Jaccard coefficient”).

 A B
T A,B  2 
A  B  A B
2
The Magic
Activity Rank

• Based on Markov chain analysis of the Activity Graph


• PageRank
– The relevance of a web page depends on the number of web pages
that link to it and the relevance of those web pages
• Activity Rank
– The relevance of a person depends on…
• The number of items she creates and edits and the relevance of
those items
• The number of people who connect with her, and the relevance
of those people
– The relevance of an item depends on…
• The relevance of its author
• The number of people who work with and like it, and the
relevance of those people
Information Flow
Enterprise Recommendation Activity Graph
Applications Service Engine

Task Schedule UI
Flows
Rank Engine
OpenUsage Client Query API
CFE
URL QRPP
Analytics Collector Gathering Engine
Security QRPP
Event Analytics Provider
Collector Event Store QRPP

Analytics Activity Graph


Schema Schema

Search Engines
Information Flow - Activity Time
Enterprise Recommendation Activity Graph
Applications Service Engine

Task Schedule UI
Flows
Rank Engine
OpenUsage Client Query API
CFE
URL QRPP
Analytics Collector Gathering Engine
Security QRPP
Event Analytics Provider
Collector Event Store QRPP

Analytics Activity Graph


Schema Schema

Search Engines

• WebCenter service writes to the Activity Stream


• OpenUsage client sends UDP unicast messages to collector
• Collector writes to the Analytics DB
Information Flow – Engine Runtime
Enterprise Recommendation Activity Graph
Applications Service Engine

Task Schedule UI
Flows
Rank Engine
OpenUsage Client Query API
CFE
URL QRPP
Analytics Collector Gathering Engine
Security QRPP
Event Analytics Provider
Collector Event Store QRPP

Analytics Activity Graph


Schema Schema

Search Engines

• Gather from Analytics and possibly other providers


• Condense and write to AG relation store
• Analyze relations to produce ranks and similarity scores
• Store results in our schema and search indexes
Information Flow – Application Runtime
Enterprise Recommendation Activity Graph
Applications Service Engine

Task Schedule UI
Flows
Rank Engine
OpenUsage Client Query API
CFE
URL QRPP
Analytics Collector Gathering Engine
Security QRPP
Event Analytics Provider
Collector Event Store QRPP

Analytics Activity Graph


Schema Schema

Search Engines

• Query scores from our schema and search engines


• Decorate and filter results using Query Result Post-Processors
(QRPPs)
Information Flow
Enterprise Recommendation Activity Graph
Applications Service Engine

Task Schedule UI
Flows
Rank Engine
OpenUsage Client Query API
CFE
URL QRPP
Analytics Collector Gathering Engine
Security QRPP
Event Analytics Provider
Collector Event Store QRPP

Analytics Activity Graph


Schema Schema

Search Engines

You might also like