Section: New Results

Linked Justifications

Participants : Rakebul Hasan, Fabien Gandon.

Semantic Web applications use inferential capabilities and distributed data in their reasoning. Users often find it difficult to understand how these applications produce their results. Hence, users often do not trust the results these applications produce. Explanation-aware Semantic Web applications provide explanations of their reasoning. Explanations enable users to better understand reasoning of these application. Users can use this additional information about reasoning to make their trust decisions.

The emergence of Linked Data offers opportunities for large-scale reasoning over heterogeneous and distributed data. Explaining reasoning over Linked Data requires explaining how these distributed data were produced. Publishing also the explanation related metadata as Linked Data enables such explanations. Justifications are metadata about how a given piece datum is obtained. We introduce the concept of Linked Justifications and provide guidelines to publish justifications as Linked Data in [67] . We published the Ratio4TA (http://ns.inria.fr/ratio4ta/ ) (interlinked justifications for triple assertions) vocabulary to describe justifications. Ratio4TA extends W3C PROV Ontology(http://www.w3.org/TR/prov-o/ ) to promote interoperability.

In [89] , [66] , we analyze the existing explanation-aware Semantic Web systems. The existing systems inherit explanation features from explanation-aware expert systems. These explanations are targeted to expert users, such as knowledge engineers, with detailed information about all the execution steps of reasoners of these applications. Unlike the expert systems, users of Semantic Web applications have diverse background - from expert knowledge engineers who are interested in every details of the reasoning, to regular users who do not have any background in reasoning, logic, or ontologies. These non-expert users might feel overwhelmed with all the execution details of reasoners. To address this issue, we propose summarized and relevant explanations to users. Users can specify their explanation goals - types of information they are interested in. We take into consideration the explanation goals when we present explanations and summarize explanations. We use centrality and similarity matrices to summarize and provide relevant explanations.