semantic integration of data - ics-forthusers.ics.forth.gr/~tzitzik/seminars/2016_06_27... ·...
TRANSCRIPT
Semantic Integration of Data Yannis Tzitzikas Computer Science Department, University of Crete, GREECE & Information Systems Laboratory (ISL) Institute of Computer Science (ICS) Foundation for Research and Technology – Hellas (FORTH)
ITN-DCH summer school 2016 (@CGI'16), Heraklion, Crete, Greece, June 27, 2016
Outline
• Motivation
• Requirements
• Case Study: Marine Species Data
• Challenges and Conclusions
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Time plan: 25’ presentation, 5’ questions and discussion
Slides: will be publicly available after the school
2
Motivation
Huge amounts of data are available and this amount constantly increases.
Almost everyone produces data (and everything will produce data).
Almost everyone needs data (and everything will need data).
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Motivation
However data and information is not integrated.
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Motivation
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Hundreds or thousands of CKAN catalogs each containing hundreds
or thousands of datasets
However data and information is not integrated.
Motivation In several domains and applications one has to fetch and assemble pieces of information coming from more than one sources for being able to answer complex queries (that are not answerable by individual sources) or analyze the integrated data. This important for science but also for our daily life.
This is true in science in general
Biodiversity domain
Cultural Domain
E-Government
Science in general
…
Personal data
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 6
50%-90% of time for data collection and cleaning
It has been written that
Data scientists spend from 50 percent to 80 percent of their time in collecting and preparing unruly digital data, before it can be explored for useful nuggets.
If you’re trying to reconcile a lot of sources of data that you don’t control it can take 80% of your time
One-Third of BI Pros Spend Up to 90% of Time Cleaning Data
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Indicative Complex Queries
Marine Domain
Given the scientific name of a species (say Thunnus Albacares), find the ecosystems, waterareas and countries that this species is native to, and the common names that are used for this species in each of the countries, as well as their commercial codes
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 8
Thunnus Albacares
El Greco
Cultural Domain Give me all paintings of El Greco that are now
exhibited in Greece and their location , as well as all articles or books about these paintings between 2000 and 2016.
Give me the paintings of El Greco referring to persons that were born between 0 and 300 BC.
Give me all events related to El Greco that will take place this month in Heraklion.
Why integration is difficult?
Datasets are kept or produced by different organizations in different formats, models, locations, systems.
The same real world entities or relationships are referred with different names and in different natural languages (natural languages have synonyms and homonyms)
Datasets usually contain complementary information
Datasets can contain erroneous or contradicting information
Datasets about the same domain may follow different conceptualizations of the domain
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
… names
348 common names in 82 different languages!
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Thunnus Albacares
ed Pa'ak Pukeu Sisek kuneng Geelvin-tuna Geelvin-tuna Tuna Tambakol Gubad Jaydher Kababa Shak zoor Tuna sirip kuning Rambayan Tambakol Bangkulis Bankulis
Bronsehan Buyo Kikyawon Paranganon Manguro O'maguro Tag-hu Taguw Taguw peras Taguw tangir Barelis Bariles Barilis Carao Karaw Pak-an Pala-pala Panit Panitto Pirit
Tulingan Kacho Bariles Karaw Panit M'Bassi Mbasi bankudi Mibassi mibankundri Thon a nageoires jaunes Thon jaune Ton zonn Z'ailes jaunes Albacora Atum olede Chefarote
Rabo-seco Gulfinnet tun Gulfinnet tunfisk Bariles Bugo Karaw Geelvintonijn 'Fin Albacore Allison tuna Allison tuna Allison tuna Allison's tuna Atlantic yellowfin tuna Autumn
albacore Long fin tunny Longfin Pacific long-tailed tuna Tuna Tuna Yellow fin tuna Yellow tunny Yellow-fin tuna Yellow-fin tuna Yellow-fin tuna Yellow-fin tunny Yellowfin
Yellowfin Yellowfin tuna Yellowfinned albacore Kulduim-tuun Gegu Tuna Yatu Yatunitoga Keltaevatonnikala Albacore Gegu Grand fouet Guegou Thon Thon a nageoires jaunes
Thon jaune Thon rouge Atu igu mera Albacore Gelbflossen-Thunfisch Gelbflossenthun Tonnos macrypteros Gedar Gedara Ahi Kahauli Kanana Maha'o Palaha Shibi Bantala-
an Panit Oriles Tambakul Tonno albacora Tonno monaco Tunnu monicu Tiklaw Vahuyo Kihada Panit Baewe Baibo Baiura Te baewe Te baibo Te bairera Te baitaba Te
ingamea Te ingimea Te inginea Bokado Olwol Malaguno Tambakol Gantarangang Lamatra Aya Aya tuna Bakulan Gelang kawung Kayu Tongkol Tuna Tuna ekor kuning Tuna
sirip kuning Poovan-choora Kannali-mas Bariles Panit Bugudi Gedar Kuppa Pimp Bwebwe Tetena keketina Vahakula Albacore Albakor To'uo Balang kuni Ghidar Albakora
Tunczyk zoltopletwy a. albakora Albacora Albacora Albacora Albacora Albacora Albacora da laje Albacora de lage Albacora-cachorra Albacora-da-lage Albacora-de-laje
Albacora-lage Albacora-lajeira Alvacor Alvacora Alvacora-lajeira Atum Atum Atum albacora Atum albacora Atum albacora Atum rabil Atum-albacora Atum-amarelo Atum-de-
barbatana-amarela Atum-de-galha-a-re Atum-galha-amarela Galha a re Ielofino Peixe de galha a re Peixe-de-galha-a-re Peixinho da ilho Rabao Rabil Rabo-seco Albacora Ton
galben Albacor Tikhookeanskij zheltoperyj tunets Zheltokhvostyj tunets Asiasi Gaogo Ta'uo To'uo Tuna zutoperka Zutorepi tunj As geddi kelawalla Howalla Howalla Kelawalla
Kelawalla Pihatu kelawalla Yajdar-baal-cagaar Albacora Albacora Albacora aleta amarilla Aleta amarilla Aleta amarilla Atun aleta amarilla Atun aleta amarilla Atun aleta amarilla
Atun aleta amarilla Atun aleta amarilla Atun de aleta amarilla Atun de aleta amarilla Atun de aleta amarilla Atun de aleta amarilla Atun de aleta amarilla Atun de aletas amarillas
Rabil Rabil Rabil Rabil Bariles Jodari Albacora Gulfenad tonfisk Albakora Badla-an Barilis Buyo Tambakol A'ahi A'ahi 'oputea A'ahi 'oputi'i A'ahi hae A'ahi mapepe A'ahi maueue
A'ahi patao A'ahi tari'a'uri A'ahi tatumu A'ahi teaamu A'ahi tiamatau A'ahi vere Otara Kelavai Soccer Soccer Tekuu Atutaoa Kahikahi Kakahi Kakahi/lalavalu Takuo Takuo
Kahikahi Kakahi Sari kanat orkinos Sar?kanatorkinoz bal?g? Sar?kanatton bal?g? Te kasi Ca bo vang Ca Ng? vay vang Nkaba Badla-an Balarito Malalag Painit Panit Baliling
Panit Tiwna melyn Doullou-doullou Ouakhandar Wakhandor Waxandor Wockhandor
… names
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
argentina
…. complementary views
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
dataset
reality
dataset
dataset dataset
dataset
dataset
…different conceptualizations
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
reality
General Requirements or Tasks related to Information Integration Dataset discovery
Dataset selection (or sub-dataset selection)
Dataset access and query
Fetch and transformation of data
Data and dataset linking
Data cleaning
Data completion (through context, inference, prediction or other methods)
Management of data provenance
Measuring and testing the quality of datasets (especially the integrated)
Management (and understanding) of the evolution of datasets
Monitoring, production of overviews, visualization of datasets
Interactive browsing and exploration of datasets
Data summarization, preservation
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
focus
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Outline
• Main approaches for integration
• The notion of Semantic Warehouse
• Case Study: Integrating information about marine species • The role Top-Level Ontologies
• Automating the process
• Measuring the quality of semantic integration
• Provenance Issues
• Exploitation
Main approaches for Integration
In general there are two main approaches for integration
Warehouse approach (materialized integration)
• Design Phase:
• The underlying sources (and their parts) have to be selected
• Creation Phase:
• Process for getting and creating the warehouse
• Maintenance Phase:
• Ability to create the warehouse from scratch, and/or ability to update parts of it
• Mappings are exploited to extract information from data sources, to transform it to the target model and then to store it at the central repository
Mediator approach (virtual integration)
• The mediator receives a query formulated in terms of the unified model/schema. The mappings are used to enable query translation. The derived sub-queries are sent to the wrappers of the individual sources, which transform them into queries over the underlying sources. The results of these sub-queries are sent back to the mediator where they are assembled to form the final answer
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 16
Main approaches for integration (cont.)
Warehouse • Benefit: Flexibility in transformation
logic (including ability to curate and fix problems)
• Benefit: Decoupling of the release management of the integrated resource from the management cycles of the underlying sources
• Benefit: Decoupling of access load from the underlying sources.
• Benefit: Faster responses (in query answering but also in other tasks, e.g. if one wants to use it for applying an entity matching technique).
• Shortcomings You have to pay the cost for hosting the warehouse. You have to refresh periodically the warehouse
Mediator • Benefit: One advantage (but in some
cases disadvantage) of virtual integration is the real-time reflection of source updates in integrated access
• Comment: The higher complexity of the system (and the quality of service demands on the sources) is only justified if immediate access to updates is indeed required.
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 17
Case Study: Marine Species
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 18
Context: iMarine
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Id: An FP7 Research Infrastructure Project (2011-2014)
Final goal: launch an initiative aimed at establishing and operating an e-infrastructure supporting the principles of the Ecosystem Approach to fisheries management and conservation of marine living resources.
Partners:
19
Continuation in BlueBRIDGE
BlueBRIDGE (Building Research environments for fostering Innovation, Decision making, Governance and Education to support Blue growth), H2020-EINFRA-2015-1
Sept. 2015- Feb. 2018
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Marine Information: in several sources
WoRMS: World Register of Marine Species
Registers more than 200K species
ECOSCOPE- A Knowledge Base About Marine Ecosystems (IRD, France)
FLOD (Fisheries Linked Data) of
Food and Agriculture Organization (FAO) of the United Nations
FishBase: Probably the largest and most extensively
accessed online database
of fish species.
DBpedia
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 21
Marine Information: in several sources
Taxonomic information
Ecosystem information (e.g. which fish eats which fish)
Commercial codes
General information, occurrence data, including information from other sources
General information, figures
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Storing
complementary information
22
Marine Information: in several sources
Web services (SOAP/WSDL)
RDF + OWL files
SPARQL Endpoint
Relational Database
SPARQL Endpoint
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Accessed through
different technologies
23
..
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
How to integrate
24
Scope Control (how to control it?)
We use the notion of competency queries.
A competency query is a query that is useful for the community at hand, e.g. for a human member , or for building applications for that domain
Indicative competency queries for our running example:
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 25
Materialization or Mediation?
In both cases we need a unified model/schema
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 26
The Top Level Ontology: MarineTLO MarineTLO aims at being a global core model that – provides a common, agreed-upon and understanding of the concepts and
relationships holding in the marine domain to enable knowledge sharing, information exchanging and integration between heterogeneous sources
– covers with suitable abstractions the marine domain to enable the most fundamental queries, can be extended to any level of detail on demand, and
– allows data originating from distinct sources to be adequately mapped and integrated
• MarineTLO is not supposed to be the single ontology covering the entirety of what exists
Benefits:
reduced effort for improving and evolving : the focus is given on one model, rather than many (the results are beneficial for the entire community)
reduced effort for constructing mappings: this approach avoids the inevitable combinatorial explosion and complexities that results from pair-wise mappings between individual metadata formats and/or ontologies
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 27
MarineTLO: Query capabilities
It should allow formulating the competency queries.
Indicative examples of queries that can be formulated
1.Given the scientific name of a species, find its predators with the related taxon-rank
classification and with the different codes that the organizations use to refer to them.
2. Given the scientific name of a species, find the ecosystems, waterareas and countries that this species is native to, and the common names that are used for this species in each of the countries
The MarineTLO currently contains around
90 classes and 40 properties.
More in www.ics.forth.gr/isl/MarineTLO
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 28
Materialization or Mediation?
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 29
We will focus on the materialization case
i.e on the construction and maintenance of a MarineTLO-based semantic warehouse
Semantic warehouse
..
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Integration process
30
The warehouse construction and evolution process
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Define requirements in terms
of competency queries
Fetch the data from the selected sources
(SPARQL endpoints, services, etc)
Queries
Transform and Ingest to the Warehouse
Inspect the connectivity of the
Warehouse
Formulate rules creating sameAs
relationships
Apply the rules to the warehouse
Rules for Instance Matching
sameAs triples Ingest the sameAs relationships
to the warehouse
Test and evaluate the Warehouse (using
the competency queries and the conn. metrics)
creates
Warehouse
produces
Triples
uses
uses
uses
MatWare
MatWare
MatWare
MatWare
MatWare
Expressed over
MarineTLO
31
How to connect the fetched pieces of information?
The Semantic Approach
Use URIs instead of strings
You can establish links in this way
You can avoid the problem of homonyms
Use owl:sameAs to connect equivalent URIs
Various other semantic relationships
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Linked Data is a method of publishing structured data so that it can be interlinked and become more useful through semantic queries. It builds upon standard Web technologies such as HTTP, RDF and URIs. This enables data from different sources to be connected and queried
How to link
We need Entity Matching
Both automatic methods and handcrafted rules are required
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Example: Suffix-based URI equivalence
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
http://www.dbpedia.com/Thunnus_Albacares http://www.ecoscope.com/thunnus_albacares
Thunnus_Albacares thunnus_albacares
ThunnusAlbacares thunnusalbacares
thunnusalbacares thunnusalbacares
≡
=
prefix removal
underscore removal
lower case conversion
last(u): is the string obtained by (a) getting the substring after the last "/" or "\#",
and turning the letters of the picked substring to lowercase and deleting the
underscore letters that might exist. 34
Example: Entity Matching-based URI Equivalence
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 35
Yannis Tzitzikas et al., LWDM 2014, Athens
http://www.ecoscope.com/thunnus_albacares http://www.fao.org/figis/flod/entities/codedentity/
636cdcea-c411-43ad-97ff-00c9304f5e60
Matching Rule: If an Ecoscope individual's preflabel in lower case is the same with the
attribute label of a FLOD individual then these two individuals are
the same.
Thunnus Albacares thunnus albacares
skos:preflabel label
sameAs
..
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
How to measure the quality of the warehouse?
36
Connectivity Assessment
In general Connectivity has two main aspects: Schema and Instance.
Regarding Schema Connectivity our running example uses a top level ontology (MarineTLO) and schema mappings in order to associate the fetched data with the schema of the top level ontology.
As regards Instance Connectivity one has to inspect and test the connectivity of the “draft” warehouse through the competency queries, and a number of connectivity metrics that we have defined and then formulate rules for instance matching
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 37
Why it is useful to measure Connectivity? For assessing how much the aggregated content is connected
For getting an overview of the warehouse
For quantifying the value of the warehouse (query capabilities)
o Poor connectivity affects negatively the query capabilities of the warehouse.
For making easier its monitoring after reconstruction
For measuring the contribution of each source to the warehouse, and hence deciding which sources to keep or exclude (there are already hundreds of SPARQL endpoints). Identification of redundant or unconnected sources
Connectivity Metrics Definition
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 38
Connectivity Metrics:
Increase in the average Degree
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
The average degree is increased from 18.72 to 23.92.
The average degree, of all sources is significantly
bigger than before.
Suffix canonicalization
Entity Matching
39
Connectivity Metrics: Exchanging
The metrics can also be exchanged for assisting dataset discovery or dataset selection (in a mediator-based architecture). We have extended VoID (Vocabulary of Interlinked Datasets) for representing and exchanging such metrics (VoIDWarehouse)
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 40
VoIDWarehouse
Connectivity Metrics: Exchanging (cont).
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
1. Compute of the Connectivity Metrics-Production of Matrixes
2. Describe the Connectivity Metrics with the proposed VoID extension
3. Store these triples in a separate graph space
4. Retrieve/Query these values from the warehouse using SPARQL queries
41
..
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Provenance
42
Provenance
It is important to keep the provenance of each data in the warehouse.
We have realized that the following 4 levels of provenance support are usually required:
[a] Conceptual level
[b] URIs and Values level
[c] Triple Level
[d] Query level
Level [a] can be supported by the conceptual model level. In our application context we use the MarineTLO and the transformation rules do the required transformations.
Matware offers support also for levels [b]-[d]
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 43
Provenance
a) Conceptual modeling level Example: Assignment of identifiers to species
MarineTLO models the provenance of species names, codes etc, and the Transformation rules of MatWare transform the ingested data according to this model.
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Thunnus albacares
YFT
127027
isIdentifiedBy
isIdentifiedBy
FAOCode
WoRMSCode hasCodeType
hasCodeType
44
Provenance
b) URIs and Literals : i. Adopting the namespace mechanism for URIs:
- The prefix of the URI provides information about the origin of the data.
- e.g. www.fishbase.org/entity/ecosystem#mediterannean_sea
ii. Ability to attach @Source to every literal coming from a Source:
- e.g. select scientific name and authorship of Yellow Fin Tunna
- This policy allows formulating source-centric queries in a relative simple way:
SELECT ?speciesname
WHERE {
?species tlo:has_scientific_name ?scientificname
FILTER(langMatches(lang(?scientificname), “worms"))
}
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 45
Provenance
c) Triple Level Provenance
• Store the fetched triples in a separate graphspace: FISHBASE: http://www.ics.forth.gr/isl/Fishbase
DBpedia: http://www.ics.forth.gr/isl/DBpedia
FLOD: http://www.ics.forth.gr/isl/FLOD
Ecoscope: http://www.ics.forth.gr/isl/Ecoscope
WoRMS: http://www.ics.forth.gr/isl/WoRMS
• By asking for the graph that each triple is coming from we retrieve the provenance of the data.
• This enables refreshing only one part of the warehouse
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 46
Provenance
d) Query Level Provenance • Matware offers a query rewriting functionality that exploits the contents of the
graphspaces for returning the sources that contributed to the query results (including those that contributed to the intermediate steps).
• Let q be a SPARQL query that has n parameters in the select clause and contains k triple patterns of the form (?s_i, ?p_i, ?o_i) :
SELECT {?o_1 ?o_2} WHERE {
?s_1 ?p_1 ?o_1 .
?s_2 ?p_2 ?o_2 .
?s_k ?p_k ?o_k
}
• The rewriting produces a query q’ that has n+k parameters in the select clause and each triple pattern (?si ?pi ?oi) has been replaced by: graph ?gi {?si ?pi ?oi}. Eventually the rewritten query q’ is:
SELECT {?o_1 ?o_2 ?g_1 ?g_2 ?g_k} WHERE {
graph ?g_1 {?s_1 ?p_1 ?o_1 }.
graph ?g_2 { ?s_2 ?p_2 ?o_2} .
graph ?g_3 {?s_k ?p_k ?o_k}
}
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 47
Provenance
Example of Query Level Provenance:
Query: For a scientific name of a species (e.g. Thunnus Albacares) find the FAO codes of the waterareas in which the species is native.
Query in SPARQL:
select ?faocode
where {
graph ?source1 {
ecoscope:thunnus_albacares MarineTLO:isNativeAt ?waterarea
}.
graph ?source2 {
?waterarea MarineTLO:LXrelatedIdentifierAssignment ?faocode
} }
RESULT:
?source1 ?source2
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 48
Architecture of Matware
Actions in order to create a Warehouse from scratch one should specify
the type of the repository
the names of the graphs that correspond to the different sources
URL, username and password in order to connect to the repository
Actions in order to add a new source
(a) include the fetcher class for the specific source as plug in
(b) provide the mapping files (schema mappings)
(c) include the transformer class for the specific source as a plug in
(d) provide the SILK rules as xml files Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 49
The resulted MarineTLO-based Warehouse(1/2)
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Integrated information about Thunnus albacares from different sources
50
The resulted MarineTLO-based Warehouse(2/2)
iMarine 2nd Review, September 2013,Brussels
Concepts Ecoscope FLOD WoRMS DBpedia Fishbase
Species
Scientific Names
Authorships
Common Names
Predators
Ecosystems
Countries
Water Areas
Vessels
Gears
EEZ
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 51
Evolution over Time
< some plots>
Need for visualization
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Exploitation of the Semantic Warehouse
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 53
Exploitation of the Semantic Warehouse
A) Semantic Processing of Search Results (e-infrastructure service)
B) Fact Sheet Generator (web application)
C) Species Identification Tool
D) Interactive 3D visualization
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 54
A) For Semantic Post-Processing of Search Results: The process
query terms (top-L) results
(+ metadata)
Entity Mining
Semantic Analysis
Visualization/Interaction (faceted search, entity
exploration, annotation, top-k graphs, etc.)
entities / contents
semantic data
web browsing
contents
• Grouping, • Ranking • Retrieving more
properties
MarineTLO
Warehouse
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 55
A) For Semantic Post-Processing of Search Results: Example (X-Search)
Search Results
Result of Entity
Mining
Result of textual
clustering
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
The Warehouse is used
The Warehouse is used
56
From FLOD
From DBpedia
From Ecoscope
From WoRMS
Example of the
EntityCard of Thunnus
Albacares
The Warehouse is used
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 57
A’) XSearch as a bookmarklet Dynamic annotating of entities over any Web page
Entity exploration
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
The Warehouse is used
58
B) Fact Sheet Generator & Android Application
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Fact Sheet Generator Ichthys
59
C) Species Identification Tool
Species identification through Preference-enriched Faceted Search over the semantic descriptions of fish species
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
D) Interactive 3D Visualization of Datasets
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
The metrics are exploited for producing interactive 3D visualization of datasets
This approach is general
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Integrated information about Thunnus albacares from different sources
62
Datasets about
Crete
Datasets about
Toledo
Datasets about
Art
The big picture (core concepts and relations)
Y. Marketakis and Y. Tzitzikas (FORTH), Edinburg, March 2012 63
Molecular world
and parts
Species (bio,geo)
Activities
Records Observations
from
cross reference
exemplifies
describe
use to
appear in
Samples or
Specimen
exemplifies collections
part of
databases
Publications
Simulation
Place
Time
is
about
global
indices
services
Forecasts
Complex
System
Situation
occurs in create maintain
from
mathem.
Models
based on
Products of Human Activities
Human Activities
The core conceptualization of Earth Sciences
..
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
What’s next
64
Challenges and our ongoing research
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Emphasis on
Dataset discovery, dataset recommendation, dataset selection (e.g. in mediator-based integration)
Finding all URIs of an entity
Finding all triples of an entity
More effective visualizations, monitoring, quality testing, trust estimation
Concluding Remarks
Semantic integration could boost data-intensive scientific discovery but requires tackling several challenging issues
We have discussed main requirements and challenges in designing, building, maintaining and evolving a real and operational semantic warehouse for marine resources
We have presented the process and related tools that we have developed for supporting this process with emphasis on Scope control, Connectivity assessment, Provenance, Reconstructability, Extensibility
Currently we focus on applying this approach for large number of datasets
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 66
Links (1/2)
MatWare (for automating the warehouse construction process)
• http://www.ics.forth.gr/isl/MatWare/
MarineTLO (top-level ontology)
• http://www.ics.forth.gr/isl/MarineTLO/
Semantic Warehouses MarineTLO-Warehouse: http://virtuoso.i-marine.d4science.org:8890/sparql
– also browsable through http://virtuoso.i-marine.d4science.org:8890/fct
XSearch (exploiting semantic warehouses in searching)
• http://www.ics.forth.gr/isl/X-Search/
Xlink (exploiting semantic warehouses for entity identification in texts)
• http://www.ics.forth.gr/isl/X-Link
<xsearch and Xlink>
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 67
Links (2/2)
Hippalus: Preference-enriched Faceted Search www.ics.forth.gr/isl/Hippalus
o Select a dataset from the Marine Biology domain for enacting the species identification through PFS
Interactive 3D Visualization of the LOD Cloud www.ics.forth.gr/isl/3DLod/
LODSyndesis: (measuring the commonalities in the entire LOD) www.ics.forth.gr/isl/LODsyndesis
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
References
On connectivity metrics
• Y. Tzitzikas, et al, Quantifying the Connectivity of a Semantic Warehouse, 4th International Workshop on Linked Web Data Management, LWDM'14@ EDBT'14)
• M. Mountantonakis et al, Extending VoID for Expressing the Connectivity Metrics of a Semantic Warehouse, 1st International Workshop on Dataset Profiling & Federated Search for Linked Data (PROFILES'14), ESWC'14,
• M. Mountantonakis, N. Minadakis, Y. Marketakis, P. Fafalios and Y. Tzitzikas, Quantifying the Connectivity of a Semantic Warehouse and Understanding its Evolution over Time, International Journal on Semantic Web and Information Systems (IJSWIS), (accepted for publication in 2016), will appear with DOI:
Recent work on for integrating large number of datasets
M. Mountantonakis and Y. Tzitzikas, On Measuring the Lattice of Commonalities Among Several Linked Datasets, VLDB’16, Sept 2016
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Acknowledgements
Joint work with
Michalis Mountantonakis
Nikos Minadakis
Yannis Marketakis
Pavlos Fafalios
Panagiotis Papadakos
Chryssoula Bekiari
Martin Doerr
Maria Papadaki
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16)
Thank you for your attention
Yannis Tzitzikas, ITN-DCH summer school 2016 (@CGI'16) 71