Not logged in.
Quick Search - Contribution
Contribution Details
Type | Conference Presentation |
Scope | Discipline-based scholarship |
Title | Identifying influential individuals from time-varying social interactions |
Organization Unit | |
Authors |
|
Presentation Type | speech |
Item Subtype | Original Work |
Refereed | Yes |
Status | Published in final form |
Language |
|
Event Title | Network Science |
Event Type | conference |
Event Location | Seoul, South Korea |
Event Start Date | May 30 - 2016 |
Event End Date | June 3 - 2016 |
Abstract Text | In the last years an increasing attention has been devoted to the identification of influential individuals and their use as super-spreaders of products or ideas. Most often, from the observed data researchers construct an influence network and identify the influencers as the most central individuals in this network or as the key players in the development of a dynamical process. However, in most practical situations there are at least two potential issues with this approach. First, the construction of the influence network is non-trivial as most often the influence relationships between people are not directly observable but rather aspects of their behavior. Second, even if the influence relationships were observable, the static network representation cannot capture their time-dynamical aspect. We present a new approach to identify influential individuals from time varying social interactions which does not require constructing the influence network nor modeling social influence as a dynamical process. We consider that individuals become influential due to unobserved features they posses, which we call the latent potential to influence. This potential is revealed during social interactions and acknowledged by other participants through rewards (e.g. upvotes on discussion platforms). We uncover the latent potential to influence from the observed rewards using the influence potential (IP), a novel index we introduce here. To illustrate our approach we analyze two real-world systems: a news discussion forum (CNN) and a business review platform (Yelp). In both datasets we find few influencers, which is in agreement with the existing theory. We compare the results against a null model and show that the presence of such individuals is very unlikely to have occurred by chance. We validate the approach by dividing the datasets into a training and a test set and showing that the users with the highest IP in the training set are also the most influential in the test set. |
Export | BibTeX |