Bartering Books to Beers: a Recommender System for Exchange - - PowerPoint PPT Presentation

bartering books to beers
SMART_READER_LITE
LIVE PREVIEW

Bartering Books to Beers: a Recommender System for Exchange - - PowerPoint PPT Presentation

Bartering Books to Beers: a Recommender System for Exchange Platforms Jrmie Rappaz Maria-Luiza Vladarean Julian McAuley Michele Catasta What is barter? Def. Barter is a system of exchange where goods or services


slide-1
SLIDE 1

Bartering Books to Beers:


a Recommender System for Exchange Platforms

Jérémie Rappaz† Maria-Luiza Vladarean† Julian McAuley‡ Michele Catasta†

† ‡

slide-2
SLIDE 2

What is barter?

  • Def. Barter is a system of exchange where goods or

services are directly exchanged for other goods or services without using a medium of exchange.

First written mention of barter around 100’000 B.C

slide-3
SLIDE 3

It’s hard to compete with money

Need of a double coincidence No common measure of value

slide-4
SLIDE 4

No common measure of value

Solution: Online bartering platforms are specialized 20’000 daily visitors 250’000 registered users 18’000 registered users

slide-5
SLIDE 5

Double coincidence of wants

Potential transactions Tom’s wish list

Heineken Duvel Brewdog Duvel Chimay Heineken Tom Jack Jane Rick

Jack’s wish list Jane’s wish list Rick’s wish list

Typical setting

slide-6
SLIDE 6

Double coincidence of wants

Solution: Matching? 85K active users 2M items

slide-7
SLIDE 7

Double coincidence of wants

Solution: Matching? 85K active users 2M items

  • nly 0.2% of users

have at least one swapping partner

slide-8
SLIDE 8

Positive signals: wish-list + past transactions

ˆ yuj,ul,ik = pT

uj qik +

Predictor - Matrix Factorization

Users Items

n x m n x k k x m

>

Unidirectional interest

slide-9
SLIDE 9

Make recommendations for one user but take into account reciprocal interest

Predictor - Bidirectionality

ˆ yuj,im,ul,ik = f(ˆ yujulim, ˆ yulujik) = 1 2(ˆ yujulim + ˆ yulujik)

>

slide-10
SLIDE 10

models a bias from one user to another.

Predictor - Social Bias

ˆ yuj,ul,ik = pT

uj qik + social bias

z }| { sujul +

users’ behavi S ∈ R|U|×|U|

++ + Some pairs of users perform recurring trades.

slide-11
SLIDE 11

Discard users/items that have been inactive for a long period

Predictor - Temporal Dynamics

2005 2006 2008 2009 2010 2012 1 2 3 4 5

ˆ yuj,ul,ik = pT

uj qik + social bias

z }| { sujul + τuj δ(t; ¯ tuj ) + τikδ(t; ¯ tik) | {z }

temporal dynamics

slide-12
SLIDE 12

Experiment

Observed trade Negative sample

>

Bayesian Personalized Ranking (BPR) 


see Rendle 2009

Maximizes AUC with positive examples only

slide-13
SLIDE 13

Results

AUC

MF MF+B MF+B+S MF+B+T MF+ALL

Bookmooch 0.758 0.798 0.849 0.938 0.958 Gameswap 0.790 0.842 0.863 0.890 0.903 Ratebeer 0.824 0.892 0.962 0.969 0.983

T S B

Bidirectionality Social bias Temporal dynamics

slide-14
SLIDE 14

Contribution & Conclusion

  • Reciprocal interest model for bartering

recommendation.

  • 3 new datasets extracted from online bartering

platforms.

  • Improving recommendations with social and

temporal information.

Powered by SIGIR Travel Grant