Modeling and Combining Evidence Provided by Document Relationships Using Probabilistic Argumentation Systems

Abstract

Previous research has shown that hypertext links may be a useful source of evidence for document contents and relevance, but these evidence are rather difficult to represent and combine. This paper introduces a new and sound model for representing and handling hypertext links in IR. This model is based on "Probabilistic Argumentation Systems", a technique which combines logic and probability to handle uncertainty. Some experiments show the potential of this approach, which requires no ad-hoc parameters.