| Summary: | Motivated by the popularity of marketplace applications over social net
works, we study optimal recommendation algorithms for online exchange
markets. Examples of such markets include peerflix. corn and readitswapit.co.uk.
We model these markets as a social network in which each user has two as
sociated lists: The item list, i.e, the set of items the user is willing to give
away, and the wish list, i.e., the set of items the user is interested in receiv
ing. A transaction involves a user giving an item to another user. Users are
motivated to engage in transactions in expectation of realizing their wishes.
Wishes may be realized by a pair of users swapping items corresponding to
each otherâs wishes, but more generally by means of users exchanging items
through a cycle, where each user gives an item to the next user in a cycle,
in accordance with the receiving userâs wishes.
In this thesis, we first consider the problem of how to efficiently gener
ate recommendations for item exchange cycles in an online market social
network. We consider deterministic and probabilistic models and show that
under both models, the problem of determining an optimal set of recommen
dations that maximizes the expected value of items exchanged is NP-hard
and develop efficient approximation algorithms for both models. Next, we
study exchange markets over time and try to optimize usersâ waiting times,
and fairness whereby fairness we mean: give higher priority to users who
contribute more to the system in addition to maximizing expected value. We
show that by introducing the concept of points, average waiting time can be
improved by a large factor. By designing a credit system, we try to maxi
mize fairness in the system. We show not only is the fairness optimization
problem NP-hard, but also inapproximable within any multiplicative factor.
We propose two heuristic algorithms, one of which is based on rounding
the solution to a linear programming relaxation and the other is a greedy
algorithm. For both the one-shot market and the overtime market studied
in this thesis, we conduct a comprehensive set of experiments, and explore
the performance and also scalability of the proposed algorithms. Our experiments suggest that the performance of our algorithms in practice could
be much better than the worst-case performance guarantee factors.
|
|---|
