The University of Melbourne
SWEN90004: Modelling Complex Software Systems Some solutions to Workshop Cx.05: Putting it all together
The intention of this tutorial exercise was primarily to stimulate discussion about the questions, rather than arrive at a particular solution. Below I have listed out some notes on the various components of the question; however, these should by no means be considered exhaustive or exclusive (ie, other correct responses are possible).
Hopefully this is of some use to those who didn’t make it to a tutorial though! 1. Getting started:
Which properties of a complex system does the “product adoption system” exhibit?
distributed system: of users / customers
decentralised: users make decisions based on local knowledge
interaction between users: communication about products services emergent behaviour: uptake of particular products / services
Who could be interested in a model of product adoption?
brands selling those particular products / services
owners of social networking platforms on which this type of marketing occurs
What specific outputs would end-users of such a model be interested in?
which products are likely to succeed
what are the factors associated with successful vs unsuccessful products which users have greatest influence in driving uptake
What real world data could you use to validate a model of product adoption?
historic data on product uptake;
historic data on product value (eg share price of company)
if this could be inferred from social media data, could also be linked with commu- nication about that product and patterns of connection amongst users
2. ODE model:
For which scenarios would this model be useful?
where we have macro level behaviour to fit to, eg size of user / potential-user populations
What are some potential shortcomings of this model?
1
doesn’t consider network effects: that some individuals may be far more influential than others
doesn’t consider competition between alternative products (eg VHS vs Betamax— if anyone knows what that refers to anymore?! Facebook vs Google+)
3. Agent-based model:
What are the advantages of the Agent-based model over the ODE model?
could capture heterogeneity in (a) individual preference/behaviour; (b) social net- work structure
Disadvantages?
more complex to implement / explore; more data required to calibrate (potentially) How would you design this model?
discuss user characteristics with stakeholders (eg companies / social media plat- forms)
explore sources of data on individual level user behaviour / network structure What are the agents?
users (potentially others, if we want to incorporate multiple products / companies into model)
What are the interactions?
communication between users
Think about some of the key questions raised throughout the course about space, time, information, state updating, interaction, decision-making, emergence and experimenta- tion.
space: could be spatial structure (grid), but network probably makes more sense
time: synchronous / asynchronous: think generally about implementation decisions
that arise, what effect they may have
information: how do agents display / observe preferences about products? how “observable” is this information?
state updating: how do agents update their preferences? balancing intrinsic pref- erence against social norms (what neighbours are doing?)
decision making: (as with state updating)
emergence: how to measure / detect patterns in uptake that occur
experimentation: baseline parameter sweeps
possible questions: how could advertising campaigns be crafted to take advan- tage of network effects? eg: targeting influencers, making use of deliberatively “provocative” advertising (ie, that will be shared, even if raising negative opinion), etc.
2