Representativeness of Knowledge Bases with the Generalized Benford’s Law

Abstract

Knowledge bases (KBs) such as DBpedia, Wikidata, and YAGO contain a huge number of entities and facts. Several recent works induce rules or calculate statistics on these KBs. Most of these methods are based on the assumption that the data is a representative sample of the studied universe. Unfortunately, KBs are biased because they are built from crowdsourcing and opportunistic agglomeration of available databases. This paper aims at approximating the representativeness of a relation within a knowledge base. For this, we use the generalized Benford's law, which indicates the distribution expected by the facts of a relation. We then compute the minimum number of facts that have to be added in order to make the KB representative of the real world. Experiments show that our unsupervised method applies to a large number of relations. For numerical relations where ground truths exist, the estimated representativeness proves to be a reliable indicator.

Anytime Large-Scale Analytics of Linked Open Data

“…The query α FSD then calculates for each property the distribution of the first significant digits of the fact number per object. This query is particularly useful for estimating the representativeness of a knowledge base by exploiting Benford's law [34]. We will use this query in Section 5 to evaluate the representativeness of the LOD cloud.…”

“…This query yields, for each property, a distribution over the frequency of the first significant digit of the number of objects per subject. We used the method proposed in [34] to convert this distribution into a score between 0 and 1 that measures the "representativeness" of the triplestores. A score of 1 means that the data is representative of the distribution in the real world (see [34] for details).…”

“…We used the method proposed in [34] to convert this distribution into a score between 0 and 1 that measures the "representativeness" of the triplestores. A score of 1 means that the data is representative of the distribution in the real world (see [34] for details). We also computed the proportion of the LOD cloud that conforms to Benford's law, and the number of distinct properties that are stored, and that are involved in the calculation of the representativeness.…”

“…As a side result, our approach estimates that the representativeness of the LOD cloud is 48.7%, which concerns 24.2% of the LOD cloud. From these numbers, we can estimate [34] that at least 13.1 billion triples are missing from the LOD cloud in order for it to be a representative sample of reality.…”

“…Several statistical indicators have been proposed to monitor the content and the quality of the data: Mapping methods [3,11,29] provide statistical indicators to summarize the property and class usage and the links between them. Other indicators evaluate the completeness of the data [16,24] or the representativeness of the properties [34]. However, the increase in volume that makes these indicators more necessary also makes them harder to compute.…”