Logical Detection of Invalid SameAs Statements in RDF Data

Laura Papaleo, Nathalie Pernelle, Fatiha Saïs, Cyril Dumont

Abstract

In the Linked Open Data (LOD) ecosystem, the owl:sameAs predicate is essential for interlinking datasets, yet it is frequently misused or erroneously generated by automated tools. This paper presents a logical framework designed to detect and invalidate incorrect identity links. By constructing "contextual graphs" that represent the local neighborhood of linked resources and applying reasoning based on functional and locally complete properties, the proposed method identifies logical contradictions. Experimental results on the OAEI Person-Restaurants dataset show that this approach significantly enhances the precision of existing data-linking methods, in some cases achieving perfect precision.

Introduction

The Semantic Web relies on the owl:sameAs predicate to assert that two different URIs refer to the same real-world entity. However, the integrity of the LOD cloud is currently threatened by an "identity crisis." Automated linking tools often produce noisy data, and human contributors frequently use sameAs to denote similarity (e.g., linking different editions of a book) rather than strict logical identity. These erroneous links propagate errors during automated reasoning, leading to data fusion failures and incorrect inferences. This research addresses the challenge of identifying these invalid links by leveraging the descriptive data already present in RDF graphs.

Methodology

The authors propose a logical approach that focuses on the "local" neighborhood of resources rather than global network analysis. The process involves three primary stages:

  1. Contextual Graph Construction: For any two resources ($x$ and $y$) linked by a sameAs statement, the system builds a contextual graph of depth $n$. This neighborhood includes the properties and values surrounding each resource.
  2. Property Selection and Reasoning: The method focuses on two specific property types:
  3. Functional Properties: Properties that must have a unique value (e.g., a restaurant's phone number). If $x$ and $y$ are identical, their values for these properties must be semantically equivalent.
  4. Locally Complete Properties: Properties where a source is assumed to contain all possible values (e.g., the complete list of authors for a specific publication).
  5. Logical Invalidation: Using Unit Resolution, the system identifies contradictions. For instance, if resource $x$ has a functional property with value $A$ and resource $y$ has the same property with value $B$, and $A \neq B$, the statement sameAs(x, y) is logically invalidated.
Contextual graphs representing the local neighborhood of a resource 'S'.

As illustrated in the figure above, the system explores various property paths (P0 through P4) originating from a resource 'S' to identify potential points of contradiction within the local data structure.

Evaluation and Results

The framework, referred to as the Invalidation Approach (IA), was evaluated using the OAEI 2010 Person-Restaurants (PR) dataset, a standard benchmark for instance matching. The IA was applied as a post-processing filter to three state-of-the-art linking methods. In every instance, the logical filter significantly improved precision:

Linking Method Original Precision Precision after IA
Method [27] 95.55% 98.85%
Method [24] 69.71% 95.19%
Method [31] 90.17% 100.00%

Notably, for Method [31], the IA detected every erroneous link, resulting in a perfect precision score. This demonstrates that strict logical filtering can effectively clean noisy datasets without requiring global analysis.

Key Insights

  • The Power of the Neighborhood: Most identity contradictions can be discovered by looking just one or two "hops" away from the resources in question, making local analysis highly efficient.
  • Strictness as a Filter: By adhering to the strict mathematical definition of owl:sameAs, the authors transform the common misuse of the predicate into a diagnostic tool for data cleaning.
  • Semantic Equality: The method recognizes that syntactic differences (e.g., "St." vs "Street") do not necessarily imply logical difference, suggesting that literal values should be clustered or normalized before reasoning.

Limitations and Future Work

While effective, the method depends on the availability of metadata regarding functional or locally complete properties, which is not always present in LOD ontologies. Furthermore, the current prototype may flag links as incorrect due to simple typos in the data. Future research will focus on integrating fuzzy string matching into the logical resolution process and exploring the "re-qualification" of invalidated links into weaker similarity predicates, such as sameAbsObject, to preserve useful but non-identical relationships.

Summary updated

Abstract

In the last years, thanks to the standardization of Semantic Web technologies, we are experiencing an unprecedented production of data, published online as Linked Data. In this context, when a typed link is instantiated between two different resources referring to the same real world entity, the usage of owl:sameAs is generally predominant. However, recent research discussions have shown issues in the use of owl:sameAs. Problems arise both in cases in which sameAs is automatically discovered by a data linking tool erroneously, or when users declare it but meaning something less ’strict’ than the semantics defined by OWL. In this work, we discuss further this issue and we present a method for logically detect invalid sameAs statements under specific circumstances. We report our experimental results, performed on OAEI datasets, to prove that the approach is promising.
BibTeX
@inproceedings{papaleo2014logical,
  author = {Laura Papaleo and Nathalie Pernelle and Fatiha Saïs and Cyril Dumont},
  title = {{Logical Detection of Invalid SameAs Statements in RDF Data}},
  booktitle = {EKAW},
  year = {2014},
  doi = {10.1007/978-3-319-13704-9\_29},
}
Figure