Chapter 13 - Simulating Social Media Data Generating Mechanism
Chapter 13 - Simulating Social Media Data Generating Mechanism
SOCIAL MEDIA
DATA
GENERATING
MECHANISMS
Contents
• Introduction
• Modeling Diffusion Of Products Within A Community
• Extending The Base Model
• Simulating The Model Using An Agent-based Model
• Initializing The ABM
• Stress Testing Input Assumptions
• Application To Diffusion Of Information On Social
Networking Websites
• A Model For CE On Social Media
• Evolution Of Consumer Engagement Over Time
• Stress Testing Assumptions
Section I
Introduction
Simulation
• Simulations are widely used in Marketing, Sociology and epidemiology
• In marketing, simulation studies are frequently performed to determine the relative relevance of
variables that are thought to promote product acceptability
• In ABM, we look at the several ways customers might be exposed to sponsored content on social
media, as well as how their interactions with that content can vary depending on these aspects
Section II
Modeling
Diffusion of
Products Within a
Community
Modeling Diffusion of Products Within a Community
Context
• Assume we're launching a new coffee variety aimed at the North Indian market.
• To fully comprehend this, we must convert the situation into an Adoption Model and simulate
the spreading process.
Extending the Bass Model
• According to this model, the number of adopters who have already adopted a product at a
certain period determines the pace of adoption.
• The Influence of Advertising and the Influence of word of mouth from friends can be used to
investigate variations of Classic Base Models.
Probability of Adoption =
This model can be translated into a simulated environment in which each community member has
a p, q, and a network of friends who can influence their decision.
Simulating the Model Using an Agent-based Model
• Agent-based modelling (ABM) is a strong tool for systematically analyzing and understanding
complicated, interconnected patterns like diffusion processes.
• The modeler in an Agent Based Model aggregates the outcomes of all agents in a network.
Consider the following scenario: If the segment has 1000 members and each has an average of
200 acquaintances, the probability of any two randomly picked members becoming friends is
200/1000 = 0.2.
We start with an unconnected group of ten individuals, cycle around each one, and with a
probability of 0.2, connect a link (‘wiring') between this individual and another in the community.
Histogram and CDF Plot
• We first establish an Erdos-Renyi network of 1000 members with a chance of connection between
two members of 0.2, then we gather the number of friends of each person and plot it in Histogram
and CDF.
Initializing the ABM
• The first step in running an ABM simulation is to set up all of the parameters that will be explored in the
analysis.
The first iteration of the ABM ( Assuming the same example as previous )
The initial iteration of the simulation necessitates the execution of the model for each member, which comes down
to the following steps:
Count how many adopted neighbours there are among these neighbours.
Examine whether advertising has had enough of an impact on the member's adoption status.
Examine whether a sufficient number of neighbours have signed the petition to sway the member.
If at least one of advertising or neighbours' influence is effective, change the member's adoption status to TRUE.
Evolving the simulation till saturation
• Now that we've completed the first run for each member in the example, we must continue the
process for numerous iterations, counting the number of adopted members each time.
• This repetition should happen till the target market reaches saturation
At the end of the execution, we have a vector number of adoptions, which comprises the
number of adopted members of the social network at each iteration.
We put this information into a data frame called simulation output and plot the number of
adoptions over time.
Graphical Plot
Stress Testing Input Assumptions
• We do this to see how resistant the output is to changes in the input.
• We investigate the error ranges surrounding rate of adoption predictions by include error bars in
our input parameters.
• The most typical way programming works is to execute the same lines of code over and over
again.
• Under this, Inputs are changed to test the robustness of the output.
Variation across multiple runs of the simulation
We can run our function numerous times to see if there are any differences while keeping all of
the parameters the same.
Variation in p and q
We check if the findings are steady when the values of p and q are changed.
We can estimate the difference in peak adoption rate due to possible noise in our estimates of
these inputs using this method.
Variation in Probability of Connection Between Members
Another sensitivity test involves changing the possibility of community members becoming friends
and assessing the impact on the peak adoption rate.
As the chance of friendships among network members grows, we predict market saturation to
occur more quickly, allowing each participant to be exposed to an acceptance signal more easily.
We can also see if networks with a lot of connections have a higher peak adoption rate.
Section III
Application to
Diffusion of
Information on
Social
Networking
Websites
Application to Diffusion Of Information on Social
Networking Websites
Context
• Assume you're a marketer, and customer engagement with your content on social media is
possible when the customer is a fan of your brand or when the customer's friends and fans are
interested in the content.
• The likelihood of a randomly selected user within the brand fans network seeing the post is 0.05.
• Given that this user's friends engaged with the post, the probability of a randomly selected user
inside the network being exposed to the post is 0.6.
• The likelihood of exposure is b1=0.05 when the consumer is a fan of the brand page, and b2=0.6
when all of the customer's friends engage with the brand post.
post
consumer.
Evolution of Consumer Engagement Over Time
We may use the Agent Based Model to look at how engagement has changed over time.
This data set may be found in the SNAP data sets collection.
The data set is a text file that contains a space-separated list of all the friendships in the network.
At each iteration, we keep track of the likelihood of the brand post being exposed to the consumer.
We think that once a consumer interacts with a brand post, they will not be exposed to it again.
Analyzing simulation output
We can now assess the simulation result by displaying the evolution of customer interaction and
the rate at which this engagement evolves.
As the number of parameters grows, it becomes more difficult to justify the exact values chosen for these
parameters.
Even if the values chosen for each of the inputs are appropriate, it is necessary to quantify the sensitivity of
the findings to modifications in these inputs.
However, when used correctly, Agent Based Models can be a useful planning tool for analysing the impact
of a brand's decisions in various scenarios.
Thank you