linked open data alignment and enrichment using bootstrapping based techniques
DESCRIPTION
The recent emergence of the “Linked Data” approach for publishing data represents a major step forward in realizing the original vision of a web that can “understand and satisfy the requests of people and machines to use the web content” – i.e. the Semantic Web. This new approach has resulted in the Linked Open Data (LOD) Cloud, which includes more than 70 large datasets contributed by experts belonging to diverse communities such as geography, entertainment, and life sciences. However, the current interlinks between datasets in the LOD Cloud – as we will illustrate – are too shallow to realize much of the benefits promised. If this limitation is left unaddressed, then the LOD Cloud will merely be more data that suffers from the same kinds of problems, which plague the Web of Documents, and hence the vision of the Semantic Web will fall short.This thesis presents a comprehensive solution to address these issues using a bootstrapping based approach. It showcases using bootstrapping based methods to identify and create richer relationships between LOD datasets. The BLOOMS project (http://wiki.knoesis.org/index.php/BLOOMS) and the PLATO project, both built as part of this research, have provided evidence to the feasibility and the applicability of the solution.TRANSCRIPT
1
About 22 years ago..
11 years later…
Image from Scientific American Website
3
4
5
6
Tim Berners-Lee 2006
1.Use URIs as names for things
2.Use HTTP URIs so that people can look up those names.
3.When someone looks up a URI, provide useful information, using the standards (RDF*, SPARQL)
4. Include links to other URIs. so that they can discover more things.
7
In 2006 Web of Data
8
8
Linked Open Data
• Massive collection of instance data
• Primarily connected via owl:sameAs relationship
• Excellent source of information for background knowledge
• Labeled as mainstream Semantic Web6/11/12
9
Is it really mainstream Semantic Web?
• What is the relationship between the models whose instances are being linked?
• How to do querying on LOD without knowing individual datasets?
• How to perform schema level reasoning over LOD cloud?
10
What can be done?
• Relationships are at the heart of Semantics
• LOD primarily consists of owl:sameAs links
• LOD captures instance level relationships, but lacks class level relationships.o Superclasso Subclasso Equivalence
• How to find these relationships?o Perform a matching of the LOD Ontology’s using state of the art ontology matching tools.
Linked Data Alignment and Enrichment
Proposal Defense June 11th, 2012
Prateek JainKno.e.sis Center
Wright State University, Dayton, OH
Agenda
• Motivation and Significance of this research
• Research questions and proposed solutions
• State of the current research and planned work
• Questions and comments
14th February 2012 12
Linked Open Data
14th February 2012 13
• A set of best practices for publishing and connecting structured data on the Web
• Practices have been adopted by an increasing number of data providers in the past 5 years
• Latest count is at 295 datasets with over 50 Billion triples (approx)
14
Linked Open Data 2007 (May)
Linking Open Data cloud diagram, this and subsequent pages, by Richard Cyganiak and Anja Jentzsch. http://lod-cloud.net/
15
Linked Open Data 2007 (Oct)
16
Linked Open Data 2009
17
Linked Open Data 2011
18
Linked Open DataNumber of Datasets
2011-09-19 295
2010-09-22 203
2009-07-14 95
2008-09-18 45
2007-10-08 25
2007-05-01 12
Number of triples (Sept 2011)
31,634,213,770
with 503,998,829 out-links
From http://www4.wiwiss.fu-berlin.de/lodcloud/state/
6 years of existence how many applications come to
your mind?
6/11/12 19
20
21
21
Reality…
• “We DID NOT use the entire Dbpedia or LOD. The only component of LOD which helped us with Watson was YAGO class hierarchy present in DBpedia. We had strict information gain requirements and other components honestly did not help much“
– Researcher with the Watson Team
6/11/12
Why?
23
A simple query..
“Identify congress members, who have voted “No” on pro environmental legislation in the past four years, with high-pollution industry in their congressional districts.”
But even with LOD we cannot answer this query.
24
Example: GovTrack
Bills:h3962
H.R. 3962: Affordable Health Care for America
Act
Votes:2009-887/+
people/P000197
Nancy PelosiOn Passage: H R 3962 Affordable Health Care for
America Act
Vote: 2009-887
vote:hasAction
vote:vote
dc:title
vote:hasOption
rdfs:labelAye
dc:title
vote:votedBy
name
25
Example: GeoNames
rdfs:subClassOf?
26
Our ApproachUse knowledge contributed by users
To enhance existing approaches to solve these issues:
• Ontology integration
• Detection relationships within and across datasets
• Querying multiple datasets
LOD Cloud
28
28
Circling Back
• LOD captures instance level relationships, but lacks class level relationships.o Superclasso Subclasso Equivalence
6/11/12
BLOOMS – Bootstrapping …
30
• BLOOMS - Bootstrapping-based Linked Open Data Ontology Matching System
• Developed specifically for LOD Ontologies
• Identifies schema level links between different LOD datasets
• Aligns ontologies belonging to diverse domains using diverse data sources
• Technique relies on using hierarchy in other datasets (therefore bootstrapping)
31
Existing Approaches
A survey of approaches to automatic Ontology matching by Erhard Rahm, Philip A. Bernstein in the VLDB Journal 10: 334–350 (2001)
32
LOD Ontology Alignment
• Actual Results from these techniques Nation = Menstruation, Confidence=0.9
• They perform extremely well on established benchmarks, but typically not in the wilds.
• LOD Ontology’s are of very different nature• Created by community for community.
• Emphasis on number of instances, not number of meaningful relationships.
• Require solutions beyond syntactic and structural matching.
33
Rabbit out of a hat?
• Traditional auxiliary data sources (WordNet, Upper Level Ontologies) have limited coverage.
• Community generated is noisy, but is rich in • Content
• Structure
• Has a “self healing property”
• Problems like Ontology Matching have a dimension of context associated with them.
34
Wikipedia
• The English version alone has more than 2.9 million articles
• Continually expanded by approx. 100,000 active volunteer editors
• Multiple points of view are mentioned with proper contexts
• Article creation/correction is an ongoing activity
35
Ontology Matching using Wikipedia
• On Wikipedia, categories are used to organize the entire project.
• Wikipedia's category system consists of overlapping trees.
• Simple rules for categorization
36
BLOOMS Approach – Step 1
• Pre-process the input ontology Remove property restrictions Remove individuals, properties
• Tokenize the class names Remove underscores, hyphens and other delimiters Breakdown complex class names
• example: SemanticWeb => Semantic Web
37
BLOOMS Approach – Step 2
• Identify article in Wikipedia corresponding to the concept.o Each article related to the concept indicates a sense of the usage of the
word.
• For each article found in the previous stepo Identify the Wikipedia category to which it belongs.o For each category found, find its parent categories till level 4.
• Once the “BLOOMS tree” for each of the sense of the source concept is created (Ts), utilize it for comparison with the “BLOOMS tree” of the target concepts (Tt).
38
BLOOMS Approach – Step 3• In the tree Ts, remove all nodes for which the parent node
which occurs in Tt to create Ts’.o All leaves of Ts are of level 4 or occur in Tt. o The pruned nodes do not contribute any additional new knowledge.
• Compute overlap Os between the source and target tree.o Os= n/(k-1), n = |z|, z ε Ts’ Π Tt, k= |s|, s ε Ts’
• The decision of alignment is made as follows.o For Ts ε Tc and Tt ε Td, we have Ts=Tt, then C=D.o If min{o(Ts,Tt),o(Tt,Ts)} ≥ x, then set C rdfs:subClassOf D if o(Ts,Tt) ≤ o(Tt,
Ts), and set D rdfs:subClassOf C if o(Ts, Tt) ≥ o(Tt, Ts).
39
Example
40
Evaluation Objectives
• To examine BLOOMS as a tool for the purpose of LOD ontology matching.
• To examine the ability of BLOOMS to serve as a general purpose ontology matching system.
41
41
Circling Back
• LOD primarily consists of owl:sameAs links
6/11/12
Part of Relationship Identification
43
Partonomy Identification
• Currently entities across datasets are linked using primarily the owl:sameAs relationship
• Relationships such as partonomy (part-of), and causality can allow creating even more intelligent applications such as Watson
• Approach PLATO (Part-Of relation finder on Linked Open DAta Tool)
44
PLATO Approach
• PLATO generates all possible partonomically linked pairs between the entities in the dataset. o Utilize “strongly” associated entities
• Identify the type of each entity in the pair using WordNet.o Use Class Nameso Gives the lexicographer files for the synsets
corresponding to these entities
• Use this information to determine the applicable OWL partonomy properties.o Using Winston’s taxonomy
45
Winston’s Taxonomy
46
PLATO Approach – Step 2
• PLATO generates linguistic patterns for each applicable property based on linguistic cues suggested by Winston.o Cell Wall is made of Cellulose
• Tests the lexical patterns for each entity pair in a corpus-driven manner.o Using Web as a corpus
• PLATO counts the total number of web pages that contain the patterno Parse the page and identify the occurance of pattern.
47
PLATO Approach – Step 3
• Asserts the partonomy property with strongest supporting evidenceo Cell Wall is made of Cellulose, 48o Cellulose is made of Cell Wall, 10
• PLATO also enriches the schema by generalizing from the instance level assertions.
48
Evaluation Objectives
• To examine PLATO as a tool for finding different kinds of part-of relation.
• To examine PLATO as a tool for finding part-of relation within a dataset
• To examine PLATO as a tool for finding part-of relation across dataset
49
14th February 2012 50
BLOOMS BLOOMS+ PLATO Others
2010 1. 1 paper at ISWC 2. 1 paper at OM
workshop
1. Paper at AAAI SS2. Paper at GEOS
2011 1. 1 paper at ESWC2. Workshop at ICBO
2012 1. 1 paper at ACM Hypertext
Total of 7 publications covering this research
Research Plan
14th February 2012 51
• Evaluation of BLOOMS on LOD ontologies
• Evaluation of PLATO
• Automatic classification of datasets
• Property alignment on LOD
Questions?