Sharing economy applications need to coordinate humans, each of whom may have different preferences over the provided service. Traditional approaches model this as a resource allocation problem and solve it by identifying matches between users and resources. These require knowledge of user preferences and, crucially, assume that they act deterministically or, equivalently, that each of them is expected to accept the proposed match. This assumption is unrealistic for applications like ridesharing and house sharing (like airbnb), where user coordination requires handling of the diversity and uncertainty in human behaviour. We address this shortcoming by proposing a diversity-aware recommender system that leaves the decision-power to users but still assists them in coordinating their activities. We achieve this through taxation, which indirectly modifies users’ preferences over options by imposing a penalty on them. This is applied on options that, if selected, are expected to lead to less favorable outcomes, from the perspective of the collective. The framework we used to identify the options to recommend is composed by three optimisation steps, each of which has a mixed integer linear program at its core. Using a combination of these three programs, we are also able to compute solutions that permit a good trade-off between satisfying the global goals of the collective and the individual users’ interests. We demonstrate the effectiveness of our approach with two experiments in a simulated ridesharing scenario, showing: (a) significantly better coordination results with the approach we propose, than with a set of recommendations in which taxation is not applied and each solution maximises the goal of the collective, (b) that we can propose a recommendation set to users instead of imposing them a single allocation at no loss to the collective, and (c) that our system allows for an adaptive trade-off between conflicting criteria.
For the full paper, see http://www.smart-society-project.eu/diversityawarerecommendation/
12. 1) “optimal” set of solutions
2) influence users choices
3) How to learn users preferences
Recommending a set of solutions: Problems
13. 1) “optimal” set of solutions
2) influence users choices
3) How to learn users preferences
Recommending a set of solutions: Problems
14. 1) “optimal” set of solutions
2) influence users choices
3) How to learn users preferences
Recommending a set of solutions: Problems
Andy Betty Charles
15. 1) “optimal” set of solutions
2) influence users choices
3) How to learn users preferences
Recommending a set of solutions: Problems
Andy Betty Charles
16. 1) “optimal” set of solutions
2) influence users choices
3) How to learn users preferences
Recommending a set of solutions: Problems
Andy Betty Charles
17. 1) “optimal” set of solutions
2) influence users choices
3) How to learn users preferences
Recommending a set of solutions: Problems
Andy Betty Charles
19. Single User and Collective of Users
User's utility function:
- how good a solution is for the user
- depends on user's requirements and preferences
System's utility function:
- depends on social welfare
- other factors
20. Single User and Collective of Users
User's utility function:
- how good a solution is for the user
- depends on user's requirements and preferences
System's utility function:
- depends on social welfare
- other factors
32. Set of Solutions Influence the Users Learning
IDEA: to artificially modify users' utility
AIM: all users prefer a solution that
lead to a feasible global solution
Explicit Approaches:
intervention
(possible) future reward
Implicit Approaches:
discounts
taxation
33. Set of Solutions Influence the Users Learning
IDEA: to artificially modify users' utility
AIM: all users prefer a solution that
lead to a feasible global solution
Explicit Approaches:
intervention
(possible) future reward
Implicit Approaches:
discounts
taxation
34. Set of Solutions Influence the Users Learning
IDEA: to artificially modify users' utility
AIM: all users prefer a solution that
lead to a feasible global solution
Explicit Approaches:
intervention
(possible) future reward
Implicit Approaches:
discounts
taxation
35. Set of Solutions Influence the Users Learning
IDEA: to artificially modify users' utility
AIM: all users prefer a solution that
lead to a feasible global solution
Explicit Approaches:
intervention
(possible) future reward
Implicit Approaches:
discounts
taxation
36. Set of Solutions Influence the Users Learning
IDEA: to artificially modify users' utility
AIM: all users prefer a solution that
lead to a feasible global solution
Explicit Approaches:
intervention
(possible) future reward
Implicit Approaches:
discounts
taxation
38. Set of Solutions Influence the Users Learning
1) utility model: what users like
2) response model: how users behave
3) belief update: done after each user's action
39. Set of Solutions Influence the Users Learning
MIPfirst
MIPothers
modify
users
utility
MIP*
standard
learning
techniques
40. PROBLEM:
properties guaranteed
by the MIPs may not hold
when taxation is applied
Set of Solutions Influence the Users Learning
MIPfirst
MIPothers
modify
users
utility
MIP*
standard
learning
techniques
56. Select the Set of Solutions such that
the Expected Value Of Information (EVOI)
is maximized
MIPs + Taxation + Active Learning
57. Select the Set of Solutions such that
the Expected Value Of Information (EVOI)
is maximized
MIPothers
MIPs + Taxation + Active Learning
58. Select the Set of Solutions such that
the Expected Value Of Information (EVOI)
is maximized
MIPothers
“Optimal Bayesian Recommendation Sets and Myopically Optimal Choice Query Set”
Paolo Viappiani and Craig Boutilier (NIPS 2010)
max EVOI → max Expected Utility
MIPs + Taxation + Active Learning
59. “Optimal Bayesian Recommendation Sets and Myopically Optimal Choice Query Set”
Paolo Viappiani and Craig Boutilier (NIPS 2010)
Query
Set of Options
Recommend an Option
MIPs + Taxation + Active Learning
60. “Optimal Bayesian Recommendation Sets and Myopically Optimal Choice Query Set”
Paolo Viappiani and Craig Boutilier (NIPS 2010)
Query
Set of Options
Recommend an Option
In the Ridesharing scenario:
commend a set of options
- only global solutions have
an utility for the system
- any solution may have
an utility for the system
MIPs + Taxation + Active Learning
61. “Optimal Bayesian Recommendation Sets and Myopically Optimal Choice Query Set”
Paolo Viappiani and Craig Boutilier (NIPS 2010)
Query
Set of Options
Recommend an Option
In the Ridesharing scenario:
● apply V. and B. work
commend a set of options
- only global solutions have
an utility for the system
- any solution may have
an utility for the system
same results
the result holds also in
this case
MIPs + Taxation + Active Learning
62. “Optimal Bayesian Recommendation Sets and Myopically Optimal Choice Query Set”
Paolo Viappiani and Craig Boutilier (NIPS 2010)
Query
Set of Options
Recommend an Option
In the Ridesharing scenario:
● apply V. and B. work
● recommend a set of options
- only global solutions have
an utility for the system
- any solution may have
an utility for the system
same results
the result holds also in
this case
MIPs + Taxation + Active Learning
63. “Optimal Bayesian Recommendation Sets and Myopically Optimal Choice Query Set”
Paolo Viappiani and Craig Boutilier (NIPS 2010)
Query
Set of Options
Recommend an Option
In the Ridesharing scenario:
● apply V. and B. work
● recommend a set of options
- only global solutions have
an utility for the system
- any solution may have
an utility for the system
same results
same results
the result holds also in
this case
MIPs + Taxation + Active Learning
64. Conclusion
Ridesharing
as real world application
uncertainty due to human behaviour
imposing → recommend
recommendation set + influence users + learning
Thank you - Questions?
65. Conclusion
Ridesharing
as real world application
uncertainty due to human behaviour
imposing → recommending
recommendation set + influence users + learning
Thank you - Questions?
66. Ridesharing
as real world application
uncertainty due to human behaviour
imposing → recommending
recommendation set + influence users + learning
67. Ridesharing
as real world application
uncertainty due to human behaviour
imposing → recommending
recommendation set + influence users + learning