Repository hosted by TU Delft Library

Home · Contact · About · Disclaimer ·

Development and evaluation of multi-agent models predicting Twitter trends in multiple domains

Publication files not online:

Author: Attema, T. · Maanen, P.P. van · Meeuwissen, E.
Publisher: Association for Computing Machinery, Inc
Source:Pei, J.Tang, J.Silvestri, F., IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining, ASONAM 2015, 25 August 2015 through 28 August 2015, 1133-1140
Identifier: 534858
doi: doi:10.1145/2808797.2808858
ISBN: 9781450338547
Keywords: Banking sector · Multi-agent models · Trend prediction · Twitter · Universities · Autonomous agents · Computational methods · Forecasting · Multi agent systems · Societies and institutions · Banking sectors · Multi-Agent Model · Trend prediction · Twitter · Universities · Social networking (online) · Human & Operational Modelling · PCS - Perceptual and Cognitive Systems · ELSS - Earth, Life and Social Sciences


This paper concerns multi-agent models predicting Twitter trends. We use a step-wise approach to develop a novel agent-based model with the following properties: (1) it uses individual behavior parameters for a set of Twitter users and (2) it uses a retweet graph to model the underlying social network structure of these Twitter users to predict trends. The model parameters can be optimized using empirical data. To investigate to what extend this agent-based model can predict Twitter trends, we validate the model performance on two case studies using real Twitter data: tweets on banks and tweets on universities. We furthermore compare a version of the model that only uses the retweet graph (PM1) with the model that also simulates individual behavior (PM2) for small to larger prediction time intervals. For both case studies the results show that PM1 performs better for small prediction time intervals (up to one day in the future), while PM2 performs better for larger time intervals (from a day to a week). We think this opens up the possibility to use similar models for helping organizations to extend their monitoring capabilities of social media with predictive modeling and to become more pro-active and less reactive. © 2015 ACM.