Naslov
Inference of influence in social networks

Autori
Matija Piškorec, Nino Antulov-Fantulin, Iva Miholić, Tomislav Šmuc, Mile Šikić

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
CompleNet'17 : Programme with Abstracts / - , 2017, 93-93

Skup
CompleNet'17

Mjesto i datum
Dubrovnik, Hrvatska, 21.03.2017. - 24.03.2017

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
online social network ; information spreading ; external influence in networks ; inference of influence in networks

Sažetak
We study the problem of inference of influence in social networks, and specifically differentiating between influence which is endogenous and that which is exogenous to the network. In the case of information diffusion in online social networks, endogenous (peer) influence is specified through an explicit influence model between peers, and it corresponds to the various ways users can interact and influence each other in online social networks, for example sharing or evaluating content generated by other users. On the other hand, exogenous (external) influence is specified as acting uniformly on all users regardless of the current state of their peers, and it corresponds to interaction which is not part of the online network, for example online news sources that independently share the same content. We define a likelihood function which can include wide range of peer influence models as well as external influence, and optimize it numerically to find maximum likelihood parameters. We evaluate our methodology on simulated activation cascades using several common models of peer influence - susceptible- infected (SI), exponential decay and logistic threshold model. We also perform inference on two large Facebook networks of 10175 and 6202 users where activation cascade is an act of registration to an online political survey application which happened during a period of one week. In addition to estimates of peer and external influence in network, our methodology is also able to characterize activation of each individual user as being peer or externally driven, and to identify most influential users.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo

POVEZANOST RADA