Detecting Errors in Numerical Linked Data Using Cross-Checked Outlier Detection

Abstract

Outlier detection used for identifying wrong values in data is typically applied to single datasets to search them for values of unexpected behavior. In this work, we instead propose an approach which combines the outcomes of two independent outlier detection runs to get a more reliable result and to also prevent problems arising from natural outliers which are exceptional values in the dataset but nevertheless correct. Linked Data is especially suited for the application of such an idea, since it provides large amounts of data enriched with hierarchical information and also contains explicit links between instances. In a first step, we apply outlier detection methods to the property values extracted from a single repository, using a novel approach for splitting the data into relevant subsets. For the second step, we exploit owl:sameAs links for the instances to get additional property values and perform a second outlier detection on these values. Doing so allows us to confirm or reject the assessment of a wrong value. Experiments on the DBpedia and NELL datasets demonstrate the feasibility of our approach.

Designing multidimensional cubes from warehoused data and linked open data

“…To separate the fundamental operations from the complex syntax of querying languages, the generated SQL and SPARQL queries are presented in the form of relational algebra and SPARQL algebra 7 respectively. Note that although the SPARQL algebra has not yet become a W3C standard, it is already supported by several frameworks for querying LOD: all algebraic expressions of SPARQL queries in this section are generated by Apache Jena API.…”

“…As the authors of [6] mention, the QB format is not always correctly used in some real-world application cases. To avoid bringing inaccurate information to decision-makers, error-detection methods such as those presented in [7] should be applied to a QB schema before applying the proposed algorithm.…”

Mining Cardinalities from Knowledge Bases

“…Under the assumption that KBs are likely to be noisy and incomplete [18], there exist several approaches that aim to enhance or refine their quality and completeness (see Paulheim [16] for a recent survey). Among the most relevant to our work are: [31] which applies unsupervised numerical outlier detection methods to DBpedia for detecting wrong values that are used as literal objects of a property; and [4] that builds upon [31] by identifying sub-populations of instances where the outlier detection works more accurately, and by using external datasets accessible from the owl:sameAs links. Our work differs from theirs in that they focus on missing property values, not on their cardinality or multiplicity.…”

Measuring Accuracy of Triples in Knowledge Graphs

“…There is a growing body of work on fact validation [6] [20] [10] [5]. This literature can be categorised into two groups in terms of the sources utilised: (1) approaches such as [20] and [10] using internal information (i.e.…”

“…A low confidence score is assigned to statements if no or only a few web pages support these sentences. The approaches [20] and [5] apply outlier detection methods to identify errors in numerical property values that are extracted from a data repository. The work [5] improves the prior work by lowering the influence of natural outliers.…”