Linked DataInterlinking

Diego Pessoaderp@cin.ufpe.br

REVIEWING… WEB AND LINKED DATA

World Wide Web is full of data!

“These data are formated for human consumption!”And the machines?

Linked Data mug

Linked Data Lifecycle

INTERLINKING

Interlinking - Definition

Interlinking refers to the degree to which entities that

represent the same concept are linked to each other.

Introduction to linked data and its lifecycle on the web (Auer, Sören Lehmann, Jens Ngomo, CAN Zaveri, Amrapali)

“connecting things that are somehow related”

Interlinking - DefinitionMetrics. Interlinking can be measured by

- Using network measures that calculate the interlinking degree

- Cluster coefficient

- SameAs chains

- Centrality and description richness through sameAs links.Airline dataset Spatial DatasetURI: americaairlines.com/country/America URI: dbpedia.org/page/United_StatesSameAs

Why Interlinking?“Include links to other URIs, so that they can discover more

things” 4th principle of LD (most important)

The goal of linking is to transform the Web into a platform for

data and information integration as well as for search and

querying.

Triples in Linked Data sources > 31 billions

-> Links consititute less than 5% of these triples

LINK DISCOVERY

Two categories of frameworks:

Linking on the Web of Data is a more generic and thus more complex

task, as it is not limited to finding equivalent entities in two knowledge

bases

Frameworks have been developed to address the lack of links

between the different knowledge bases on the web.

1) Ontology matching:establish links between ontologies underlying two data sources.

2) Instance matching (link discovery):discover links between instances contained in two data sources.

LINK DISCOVERYFormally…

Given Two sets S (source); T (target) of instances,a (complex) semantic similarity measure σ : S × T → [0, 1] and a threshold θ [0, 1]∈

The goal of link discovery task is to compute the set M = {(s, t), σ(s, t) ≥ θ}.

In general, the similarity function used to carry out a link discovery task

is described by using a link specification (sometimes called linkage

decision rule).

CHALLENGESTwo key challenges arise when trying to discover links between two sets of instances:

1) computational complexity of the matching task

2) selection of an appropriate link specification.

CHALLENGES1) Computational complexity of the matching task• The time complexity of a matching task can be measured

by the number of comparisons necessary to complete this task

• Reduction of the time complexity of link discovery is a key requirement to instance linking frameworks for Linked Data.

Ex.: discovering duplicate cities in Dbpedia would necessitate approximately 0.15 × 109 similarity computations.

CHALLENGES2) Selection of an appropriate link specification.

• The configuration of link discovery frameworks is usually carried out manually, in most cases simply by guessing

• Methods such as supervised and active learning can be used to guide the user in need of mapping to a suitable linking configuration for his matching task

APPROACHES TO LINK DISCOVERYCurrent frameworks for link discovery can be subdivided into two main categories:

Domain-specific

Universal

• RKBExplorer (academic purposes)• GNAT (music)

• RDF-AI (not time optimized)• LIMES (time optimized)• SILK

ACTIVE LEARNING OF LINK SPECIFICATIONSThe second challenge of Link Discovery is the time-efficient discovery of link specifications for a particular linking tasks.

Several approaches have been proposed to achieve this goal, of which most rely on genetic programming

COALA (Correlation-Aware Active Learning) approach was implemented on top of the genetic programming

CONCLUSIONS• First works on running link discovery in parallel have

shown that using massively parallel hardware such as GPUs can lead to better results that using cloud implementations even on considerably large datasets.

• Detecting the right resources for linking automatically given a hardware landscape is yet still a dream to achieve.

CURRENT CHALLENGES

• Authoring

• Extraction from structured fonts (RDBS, XML)

• Natural Language Queries

• Automatic Management of Resources for Linking

• Linked Data Visualization

• Linked Data Quality/Reliability

Main References[Book] Linked Data – Structured data on the web. David Wood, Marsha Zaidman, Luke Ruth. 2014

[Paper] Linked Data – The story so far. Berners Lee.

Linked DataInterlinking

Diego Pessoaderp@cin.ufpe.br

Thanks!