Influence Factor: Extending the PROV Model With a Quantitative Measure of Influence

A central tenet of provenance is to support the assessment of the quality, reliability, or trustworthiness of data. The World Wide Web Consortium’s (W3C) PROV provenance data model shares this goal, and provides a domain-agnostic interchange language for provenance representation. In this paper we suggest that given the PROV model as it stands, there are cases where information relating to how one entity has influenced another falls short of that required to make these assessments. In light of this, we propose a simple extension to the model to capture a quantitative measure of influence. To understand how provenance publishers use PROV to describe influence we have consulted the current Provbench datasets and evaluated the usage of the 13 sub properties ofwasInfluenced By. The findings suggest that publishers are willing to provide additional information about how an influencer affected an influencee beyond a simple wasInfluencedBy relation. In the paper, we define influence factor as a quantitative measure of influence that one PROV entity, agent, or activity has had over another and introduce influenceFactor as property to enrich any qualified influence in the PROV model. To demonstrate the use of the use of influenceFactor we have extended the Wikipedia-provenance dataset and tooling from ProvBench to capture a quantitative measure of influence between the provenance elements involved. We also briefly discuss how we have used the proposed influence factor to support the development of a probabilistic approach to information quality (IQ) assessment using Bayesian Networks.