Reasoning with Noisy Semantic Data

Based on URIs, HTTP and RDF, the Linked Data project [3] aims to expose, share and connect related data from diverse sources on the Semantic Web. Linked Open Data (LOD) is a community effort to apply the Linked Data principles to data published under open licenses. With this effort, a large number of LOD datasets have been gathered in the LOD cloud, such as DBpedia, Freebase and FOAF profiles. These datasets are connected by links such as owl:sameAs. LOD has gained rapidly progressed and is still growing constantly. Until May 2009, there are 4.7 billion RDF triples and around 142 million RDF links [3]. After that, the total has been increased to 16 billion triples in March 2010 and another 14 billion triples have been published by the AIFB according to [17]. With the ever growing LOD datasets, one problem naturally arises, that is, the generation of the data may introduce noise, thus hinders the application of the data in practice. To make the Linked Data more useful, it is important to propose approaches for dealing with noise within the data. In [6], the authors classify noise in Linked Data into three main categories: accessibility and derefencability w.r.t. URI/HTTP, syntax errors, and noise and inconsistency w.r.t. reasoning. In our work, we focus on dealing with the third category of noise, namely noise and inconsistency w.r.t. reasoning, but may also consider other categories of noise. We further consider one more noise in the logical level, that is, the logical inconsistency caused by ontology mapping.