Divesting in Socially (Ir)responsible Internet Service Providers

Abstract

Divesting in a company based on their social practices is one tactic for encouraging a company to alter their behavior. Individuals and institutions can divest in "socially irresponsible" ISPs directly by refusing to peer with or purchase transit from these ISPs, or indirectly by avoiding paths through these ISPs. While the economics of divesting in ISPs and mechanisms for exerting more control over Internet paths have been previously studied, a major challenge remains: how do individuals and institutions assess whether an ISP is socially responsible? Large institutions may have the resources to research which ISPs are socially (ir)responsible, but most individuals and institutions lack these resources, thereby limiting their ability to effect social change. We address this challenge by designing a system for automatically assessing ISPs along various dimensions of social responsibility---e.g., environmental stewardship, censorship actions, or net neutrality practices. In particular, we gather publicly accessible web pages which discuss a particular ISP and social issue and apply stance detection---a natural language processing technique for determining a subject's attitude or commitment toward an object---to compute a "socially responsibility" score for individual ISPs and dimensions of social responsibility. We construct a data set over 600 web pages relating to net neutrality and 13 ISPs of varying sizes, and show our framework achieves 71% accuracy in determining an ISP's stance.