Abstract:
We propose a new activation function (Trust) with desirable mathematical properties intended for influence dynamics in social systems. We show that that Trust, which acco...Show MoreMetadata
Abstract:
We propose a new activation function (Trust) with desirable mathematical properties intended for influence dynamics in social systems. We show that that Trust, which accounts for both a targeted user's opinion and willingness to learn, outperforms popular alternatives i.e. Sigmoid, TanH, in both controlled and uncontrolled systems across variations of network size, initial opinion, credibility, and duration. We modify the influence dynamics model first introduced by [1] to allow for continuous gradient updates and effectively shift a user's opinion in a finite number of steps. The implications of this research suggest that (i) selectivity of activation functions in a social system can drive more effective influence, (ii) it is possible to account for phenomenon such as stubbornness, skepticism, and extremism when targeting a user. We believe the results can potentially enhance applications in marketing, political campaigning, and general persuasion.
Date of Conference: 18-22 July 2021
Date Added to IEEE Xplore: 22 September 2021
ISBN Information: