Computer Science News
UCL delivers first systematic study of the network effects shaping digital reputation in P2P platforms
Three researchers from UCL Computer Science, Giacomo Livan, Fabio Caccioli, and Tomaso Aste, have shown that the reputation systems that underpin online P2P platforms (such as Uber and AirBnB) can be made unreliable for all participants by some users’ tendency to exchange ratings. The findings are published in the 14th June issue of Nature Scientific Reports: “Excess reciprocity distorts reputation in online social networks”.
The current landscape of the online digital economy is largely organized as a “platform society” of users who exchange knowledge, goods, and resources on a peer-to-peer basis. In order to trust each other, P2P platform users develop reputation scores through digital peer-review mechanisms, such as the star ratings that Airbnb hosts and guests or Uber drivers and passengers give each other after a stay or a ride. P2P platforms already occupy a prominent role in the digital economy, and they already generate billions in global revenues each year. The ability of both platforms and their users to take advantage of this economic growth will be determined by the reliability of these reputation systems.
This work makes use of Network Theory to analyse data generated by P2P platform users, and finds statistical evidence that users overwhelmingly tend to reciprocate ratings, and that this impacts aggregate reputation scores. A well-documented example is the “5 for 5” practice where Uber drivers and passengers agree on a mutual exchange of 5 star ratings to boost their star-based reputation scores. Indeed, the research verifies on a variety of platforms that this rating reciprocation makes up the bulk of ratings, making the aggregate reputation scores themselves substantially less meaningful.
In all, this work shows that the very high online reputation scores we often see in online platforms are largely due to a collective rating exchange process, which effectively prevents from being able to discriminate between high and low quality participants. As such, it undermines one of the very reasons online reputation systems were introduced.
The article's digital object identifier is 10.1038/s41598-017-03481-7.