Logo Logo
Help
Contact
Switch Language to German
Franzke, Maximilian; Bleicher, Janina; Züfle, Andreas (2016): Finding Influencers in Temporal Social Networks Using Intervention Analysis. In: Cheema, Muhammad Aamir; Zhang, Wenjie; Chang, Lijun (eds.) : Databases Theory and Applications: 27th Australasian Database Conference, ADC 2016, Sydney, NSW, September 28-29, 2016, Proceedings. Lecture Notes in Computer Science, Vol. 9877. Cham: Springer. pp. 3-16
Full text not available from 'Open Access LMU'.

Abstract

People influence, inspire and learn from each other. The result of such a latent cooperation can be observed in social networks, where interacting users are connected with each other. In previous work, the influence casted by a single individual is measured purely quantitatively, by analyzing the topology of a social network. In this work, we take a step towards analyzing the quality of an influencer. For this purpose, we analyze how attributes describing an individual change over time, and put this change in the context of the social network topology changing over time. Each social node is associated with time-series of their attribute values, such as their number of citations. For each individual, we apply the concept of intervention analysis, to identify points in time, so-called interventions, when these attributes have been affected significantly. Such interventions can be of either positive or negative type, and either short-term or long-term. For each intervention, we use the temporal social network's topology to identify candidate individuals, potentially responsible for this intervention. We use these interventions to score their influence on others. This allows us to find users, who have a significant bias towards affecting the performance of other users, in a positive or negative way. We evaluate our solution using a data set obtained from the ACM digital library - containing both a temporal collaboration network, as well as time-dependent attributes such as the number of citations. Our resulting most-positively influential researches is quite different from traditional purely quantitative metrics such as citation count and h-index.