Slides of LDOW2013 presentation, May 14th, Rio De Janeiro, Brazil We will show that semantically annotated paths lead to discovering meaningful, non-trivial relations and connections between multiple resources in large online datasets such as the Web of Data. Graph algorithms have always been key in pathfinding applications (e.g., navigation systems). They make optimal use of available computation resources to find paths in structured data. Applying these algorithms to Linked Data can facilitate the resolving of complex queries that involve the semantics of the relations between resources. In this paper, we introduce a new approach for finding paths in Linked Data that takes into account the meaning of the connections and also deals with scalability. An efficient technique combining pre-processing and indexing of datasets is used for finding paths between two resources in largedatasets within a couple of seconds. To demonstrate our approach, we have implemented a testcase using the DBpedia dataset.

1. ELIS – Multimedia Lab
Discovering Meaningful Connections between
Resources in the Web of Data
Everything is connected: behind the scenes
Laurens De Vocht
Sam Coppens, Miel Van der Sande, Ruben Verborgh, Erik Mannens, Rik Van de Walle

2. ELIS – Multimedia Lab
Barack Obama
?
Paris Barack ObamaBertrand
Delanoë
Catholic
Church
Joe Biden
mayor religion religion of vicepresident of
Query
Result
Paris
Everything is Connected

3. ELIS – Multimedia Lab
Reource
Description
Resource
Lookup
Pathfinding
Resource
Description
Everything is Connected

4. ELIS – Multimedia Lab
1. Pathfinding
2. Optimizations
3. Performance Evaluation
4. Future Work & Discussion
Overview

5. ELIS – Multimedia Lab
1. Pathfinding
2. Optimizations
3. Performance Evaluation
4. Future Work & Discussion
Overview

6. ELIS – Multimedia Lab
Pathfinding
Use of A* Algorithm
Requires:
Adjacency Matrix
Weighted links
Heuristic

7. ELIS – Multimedia Lab
Pre-processing
RDF Linked Data (NQ)
<http://dbpedia.org/resource/Aristotle>
<http://xmlns.com/foaf/0.1/name>
"Aristotle"@en
<http://en.wikipedia.org/wiki/Aristotle?oldid=494
147695#absolute-line=10> .
…
INCOMING TRIPLES
...
(Triples pointing to subject)
METADATA
http://dbpedia.org/resource/A
ristotle
(subject)
http://en.wikipedia.org/wiki/Ar
istotle?oldid=494147695#ab
solute-line=10
(context)
OUTGOING TRIPLES
<http://dbpedia.org/resource/Aristotle>
<http://xmlns.com/foaf/0.1/name>
"Aristotle"@en
...
(Triples subjects points to)

8. ELIS – Multimedia Lab
Pre-processing
SIREn
Indexing

9. ELIS – Multimedia Lab
Pathfinding
Use of A* Algorithm
Requires:
 Adjacency Matrix
Weighted links
Heuristic

10. ELIS – Multimedia Lab
Adjacency Matrix Initialisation
Paris Barack Obama
Betrand Delanoë
France
...
Joe Biden
...
United States
global set of all
resources increases
every iteration

11. ELIS – Multimedia Lab
Adjacency Matrix Initialisation
List indices correspond with row/column numbers in adjacency matrix
Generation of list of all resources

12. ELIS – Multimedia Lab
Pathfinding
Use of A* Algorithm
Requires:
 Adjacency Matrix
 Weighted links
Heuristic

13. ELIS – Multimedia Lab
Weighted Links
Weight encourage rare nodes in paths (Moore et al.)

14. ELIS – Multimedia Lab
Pathfinding
Use of A* Algorithm
Requires:
 Adjacency Matrix
 Weighted links
 Heuristic

15. ELIS – Multimedia Lab
Heuristic
Heuristic: Jaccard distance
measures dissimilarity between sample sets
FranceParis
:monument
:mayor
:language :capital

16. ELIS – Multimedia Lab
1. Pathfinding
2. Optimizations
3. Performance Evaluation
4. Future Work & Discussion
Overview

17. ELIS – Multimedia Lab
Node centrality based rank reduction
Blacklisting irrelevant links
Optimizations

18. ELIS – Multimedia Lab
1. Pathfinding
2. Optimizations
3. Performance Evaluation
4. Future Work & Discussion
Overview

19. ELIS – Multimedia Lab
HITRATE
COMPLEXITY
EXECUTION TIME
Performance Evaluation

20. ELIS – Multimedia Lab
Hitrate above 90%
Testset of 10 000 paths
among 200 popular cities, artists and countries
in DBPedia (10M entities)

21. ELIS – Multimedia Lab
Exponential space complexity

22. ELIS – Multimedia Lab
Linear time complexity

23. ELIS – Multimedia Lab
Over 60% of paths found in less than 2000ms

24. ELIS – Multimedia Lab
1. Pathfinding
2. Optimizations
3. Performance Evaluation
4. Future Work & Discussion
Overview

25. ELIS – Multimedia Lab
Synchronisation
Index <> Source Repositories/Endpoints
Performance
Further improve iterative reduction
(selection of potentially relevant entities)
Personalization
Adapt blacklist, heuristic and link weights to user preference and context
Future Work & Discussion

26. ELIS – Multimedia Lab
Optimized pathfinding for
linked data to obtain
meaningful results.
Results within a tolerable time
for users.
Conclusions
http://pathfinding.restdesc.org
http://www.everythingisconnected.be
@laurens_d_v #mmlab
laurens.devocht@ugent.be
http://slideshare.net/laurensdv
http://semweb.mmlab.be/
Contact