Résumé
The main aim of truth-finding methods is to identify the most reliable and trustworthy data among a set of facts. Since existing methods assume a single true value, they cannot deal with numerous real-world use cases in which a set of true values exists for a given fact, even for functional predicate (e.g. Picasso is born in Màlaga and in Spain). This paper studies how traditional truth-finding methods can be adapted to this setting. After introducing a new definition of true value and discussing associated implications, we propose an approach that can be used to identify true values among a set of non-conflicting claims; it takes advantage of belief functions to incorporate knowledge about value relationships in the form of a partial ordering of claimed values. By reducing the error rate up to 30% adapting classical approaches, the effectiveness and suitability of our proposal is clearly highlighted through empirical experiments performed on DBpedia.