Feature engineering from the perspective of agenda setting for predicting the success of online petitions

Abstract

This study draws on the issue expansion model and symbolism, both of which are influential concepts in the literature of public policy and agenda setting, to generate textual features for developing a predictive model of online petition success. Using a real-life dataset of an online petition platform, we show that the proposed model performs well in several important evaluation metrics when compared with benchmark models. This study offers several contributions. First, we present how to translate these concepts into textual features of petitions that can be understood by computers to improve prediction of petition success. The predictive models developed and the patterns of online petitioning identified enhance our understanding of collective actions on online petition platforms. In addition, we demonstrate that we can develop a better predictive model by adopting both supervised and unsupervised approaches of model development together with datasets that are exogenous from online petition platforms. Further examination of the predictive models in future may enable us to define vague concepts in a systematic way. On practical implications, our proposed text-mining model enables policy makers to handle a large volume of social data in a relatively objective manner. This is conducive to civic participation in e-democracy. The model may help policy makers identify potentially popular issues and prevent issue expansion at an early stage to mitigate the possible incursion of social cost. Moreover, by developing a predictive model based on our approach, citizens can compare different petition texts to determine their chances of success and post texts that have a higher predicted rate of success.