2 of 22

What is Behavior Analytics

What motivates someone to join a group?
When someone abandons social media “A”, where do they migrate to?
Can we predict box office revenues of movies based on tweets posted by individuals?
Etc……..

3 of 22

Types of Individual Behavior

User user

Person to person

User community
User entity

Friending someone

Following someone

Blocking

Playing games

4 of 22

Types of Individual Behavior

User user
User community

User to a community

User entity

Friending someone

Following someone

Blocking

Playing games

Joining/leaving a group

Becoming fan of a celebrity

5 of 22

Types of Individual Behavior

User user
User community
User entity

User to an entity in social media

Friending someone

Following someone

Blocking

Playing games

Joining/leaving a group

Becoming fan of a celebrity

Write a status

Write a blog

Upload a photo

6 of 22

What can we do?

Individual Behavior Analysis

Why are they doing this?

Individual Behavior Prediction

Are they going to do this?

7 of 22

Behavior Analysis

Act of joining a community by a user: user-community behavior
Alice and Natalie are both gonna buy a snowblower X

Joining the community of people who have X already

Analyze:

What is common between Alice and Natalie
Why are they joining this “snowblower X” community?

8 of 22

Behavior Analysis

We can have some hypothesis
One hypothesis:

Alice and Natalie are buying X because most of their close friends might use X already
True/False?
Collect data and analyze and validate!

9 of 22

Some other factors that influence users to joining a community

https://dl.acm.org/doi/pdf/10.1145/1150402.1150412

10 of 22

Behavior Analysis Methodology

Find something to observe

Group joining behavior

Buying a product

Etc…

11 of 22

Behavior Analysis Methodology

Find something to observe

Group joining behavior

Buying a product

Etc…

Extract features to use for analysis

# of Friends already in the community

12 of 22

Behavior Analysis Methodology

Find something to observe

Group joining behavior

Buying a product

Etc…

Extract features to use for analysis

# of Friends already in the community

See how the feature influences

# of Friends already in the community more 🡪 more probability of joining

Evaluate

Collect data and evaluate if that is true

13 of 22

Behavior Prediction

Will user X buy this product?
Will user Y join this group?

14 of 22

Behavior Prediction

Node neighborhood-based methods

Jaccard
Adamic-Adar
Preferential Attachment

Path Based Methods

Katz
SimRank
Rooted PageRank

15 of 22

Adamic Adar

Idea

If A and B nodes share a neighbor C
And that neighbor, C, is a ”rare” neighbor
Then A and B are more similar

16 of 22

Adamic Adar

Idea

If A and B nodes share a neighbor C
And that neighbor, C, is a ”rare” neighbor
Then A and B are more similar

Similarity between 5 and 7

17 of 22

Adamic Adar

Idea

If A and B nodes share a neighbor C
And that neighbor, C, is a ”rare” neighbor
Then A and B are more similar

Similarity between 5 and 7

x = 5, y = 7

Common neighbors between 5 and 7 : 4

u = 4

N(u) = {5,6,7} so |N(u)| = 3

A(5,7) = 1/ (log3)

18 of 22

Preferential Attachment

Idea

Nodes of higher degrees are more similar
More probability of being connected

19 of 22

Preferential Attachment

Idea

Nodes of higher degrees are more similar
More probability of being connected

Sim(x,y) = |N(x)| * |N(y)|

Sim(5,7)

N(5) = |{4,6}| = 2

N(7) = |{7}| = 1

Sim(5,7) = 2*1 = 2

20 of 22

Katz: Paths based measure

Idea

Nodes that are more similar have many paths of varying lengths connecting them together
More shorter paths: more similar
Too many longer paths: not that similar

21 of 22

Katz: Paths based measure

Suppose, ℬ, attenuation factor is 0.5

22 of 22

Katz: Paths based measure

Suppose, ℬ, attenuation factor is 0.5

Katz(A,B)

L = 1=> AB

L = 2 => ACB

L = 3 => ACDB

(0.5)¹* 1+ (0.5)² * 1 + (0.5)³ * 1 + ….