Enabling Data Analytics from Knowledge GraphsUniversidade de FortalezaFortalezaBrazil hos@edu.unifor.br
Scienti c data is being acquired in high volumes in support of studies in many knowledge areas. Regular data analytics processes make use of datasets that often lack enough knowledge to facilitate the work of data scientists. By relying on knowledge graphs (KGs), those di culties can be mitigated. This research focuses on enabling data analytics over scienti c data in light of knowledge available in KGs, providing access, based on queries, to scienti c data points in KGs to data users while making use of available knowledge to facilitate their data analytics activities.
knowledge graphsdata analyticsdata access-
{ How to successfully nd all the relevant data among massive data collections?
{ How to compare/combine two (or more) variables that measure the same
characteristic but were acquired using di erent instruments each one with
its own resolution, precision and accuracy?
{ How to allow data users, with no prior knowledge about the data, to
successfully use it and leverage new applications?
Recently, the use of Knowledge Graphs (KGs) is on the rise as a way of
building large knowledge bases as a graph structure. Those graphs aim to
represent knowledge as a series of statements as triples, in the form of subject
predicate object. Until now, KG common usages include enhancing search [
1
]
and performing A.I. tasks like Question Answering [
17
], Natural Language
Processing [
7
] and Machine Learning [
18
]. In contrast, there exists an increasing
number of approaches for building domain sciences [
2, 6, 9
], Internet of Things
(IoT) [
16
] and city [
23
] KGs (in which scienti c data are present), with the
intent of encoding provenance, context and further knowledge behind each
scienti c data point. Nevertheless, when confronted with above listed problems,
the state-of-the-art approaches do not perform well as they are either focused
on annotation [
2
], general use [
9
] or real-time querying [
16
]. As a consequence,
data users rely yet on the aforementioned scienti c datasets that usually lack
enough knowledge to facilitate their data understanding and preparation.
This Ph.D. research proposal settles itself on the problem of enabling data
analytics over scienti c data in light of knowledge available in KGs that describe
studies generating scienti c data. More speci cally, to provide access, based on
queries, to scienti c data points in KGs to data users while making use of
available knowledge to facilitate their data analytics activities. This objective poses
a number of challenges, among them:
{ Domain modeling: The development of domain ontologies for the Semantic
Web has been historically use-case driven [
4, 11
], but the data analytics
usecase has not yet been fully explored.
{ Provenance, contextual knowledge and uncertainty: Instruments
characteristics (resolution, accuracy, precision), agent interventions over them
(deployments, calibrations, con gurations) and detectors faults are
examples of what can directly a ect scienti c data values. This knowledge needs
to be tracked and explored to provide data users trustworthy results.
{ Knowledge Graph data access: Routine data tools (R, Python, Weka,
Gephi, Business Intelligence softwares etc.) used in support of data
analytics activities often expect tabular data as input, not coping properly with
Semantic Web technologies and formats.
2
Relevancy
Data preparation is estimated to take around 80% of the whole analytical pipeline
[
19
], with tasks like data understanding and data cleaning requiring a great e ort.
Hence, the actual analysis activities, which indeed extract new knowledge from
the data, are delayed and/or shortened, directly impacting outcomes quality and
projects deadlines. We aim to simplify this process by providing speci cations
and tools. Given this, we expect this research to bring straight bene ts to data
scientists and eld specialists.
Interoperability between scienti c KGs and existing non-semantic tools should
broaden the use of KGs to even more knowledge areas, as working with data
contained in it will be made easier.
3
Related work
The \Knowledge Graph" term gained popularity with the announcement1 of
The Knowledge Graph by Google in an e ort to merge Freebase[
5
], Wikipedia
and the CIA World Factbook2 augmented with their search engine's queries
and results. Since then, a number of existing projects have been categorized as
KGs. For instance, YAGO [
25
], DBpedia [
3
] and Wikidata [
26
] are free
generalpurpose KGs, while the Gene Ontology [
2
], Bio2RDF [
6
] and KnowLife [
9
] are
aimed toward life sciences. All of those approaches have focused on the problem
of encoding knowledge and KG building, not particularly concerned about how
to cope with data analytics activities.
The Graph of Things [
16
] is a proposed KG for integrating heterogeneous
IoT data sources that enables querying and visualization through an SPARQL
endpoint. This approach makes use of the SSN ontology3 to describe physical
sensing instruments and their observed data with some metadata including
sensor con guration and measured characteristic. However, the sole ways to work
with data contained in this KG is to either use its SPARQL endpoint or a
stream subscribing channel that provides continuous queries over RDF stream
data which, therefore, makes this approach not suited for data analytics, lacking
interoperability with existing data tools. The work in [
8
] describes an approach
that integrates heterogeneous data sources into an RDF KG for predictive
analysis. The presented system is capable of providing an SPARQL query interface
for preparing datasets for di erent tools in the context of predictive analysis.
There exists a number of approaches tackling data analytics challenges
related to the Smart City context using city published data. CityPulse [
22
] is a
framework that enables development of applications in support of cities, by
providing integration mechanisms for urban data streams. The ISO 37120:2014 [
14
]
is a standard that de nes 100 indicators across 17 themes that were evaluated
to be a precise way to measure a city's performance of its services and quality of
life. The themes span areas including Economy, Education, Health, and Safety.
The main goal of this standard is to provide a concise set of well-de ned global
indicators that any city can use to measure itself. Moreover, cities that adhere
to this standard are able to compare themselves and evaluate how well they are
1
https://googleblog.blogspot.com/2012/05/introducing-knowledge-graphthings-not.html
2 https://www.cia.gov/library/publications/the-world-factbook
3 https://www.w3.org/TR/vocab-ssn/
doing in comparison to others. Relying on the RDF model and making use of
the ISO standard, the PolisGnosis Project [
10
] is a nal goal of an ongoing e ort
by the University of Toronto. The project aims the following:
{ To provide a description of all the 100 ISO indicators in terms of ontologies
for the semantic web;
{ To develop an engine capable of performing analysis in order to discover root
causes of di erences concerning why indicators change over time for a given
city and why they are di erent between di erent cities.
Until the time of this writing, the PolisGnosis Project has focused largely
on the GCI Ontology engineering[
11
] as a standard to publish the ISO indicator
values.
4
Research questions
Given the identi ed challenges and limitations presented in the previous sections,
we have formulated the following research questions that we intend to answer:
Q1 Can ontologies be used to successfully bridge the knowledge gap between
acquired scienti c data and data users? If so, how?
Existing scienti c domain ontologies are not aligned with data analytics
requirements. For instance, when calculating indicators, related concepts may
suggest that a certain data point should be taken into account and these
relations are not always present because the ontology was developed for another
purpose.
Q2 Will data users and applications bene t from the use of knowledge behind
each scienti c data point?
Common search mechanisms only index dataset metadata, returning
complete datasets that may be in a plethora of di erent formats.
Q3 How to provide data access for scienti c KGs in a way that can be consumed
by routine data tools while making use of the attached data knowledge to
facilitate analytics?
Current RDF serialization formats include Turtle, JSON-LD and RDF/XML
which are not suited for most data tools, while SPARQL querying requires
previous knowledge on the ontologies used in the KG.
5
Hypotheses
Our hypotheses derive directly from the questions above:
H1 The reuse of scienti c data ontologies with proper extensions and their
alignments to domain ontologies can mitigate the loss of knowledge during data
acquisition.
H2 Providing data points together with their knowledge (e.g. provenance,
contextual knowledge) to data users and applications can facilitate data
analytics.
H3 A hybrid RDF serialization format that suits the needs of existing data tools
but also is able to convey knowledge can be used to serialize data from KGs
together with its associated metadata.
H4 A query API for scienti c KGs can also be used to output data together
with its associated metadata for facilitating data analytics.
6
Preliminary results
In [
24
], we described our rst approach at a process of data acquisition and KG
building in the context of urban mass transportation where data was produced
by GPS devices deployed on buses from the city of Fortaleza, Brazil. The built
KG was suited for metadata-driven faceted-search over the data, which enabled a
better understanding of the data contained in the KG by the explicit information
about context provenance. This work was our rst approach of putting in place
enough relevant metadata and this was accomplished by the use of our HAScO
Ontology4 (which evolved from HASNetO [
21
]) as a way of describing content
and context of the acquired data.
Following, in [
23
], we presented an operational description of a KG that
supports automatic generation of dashboards along with an indicator ontology
that supports data visualization techniques. This work extends the previous
one by providing a rst data analytics use-case where data in KG is used to
produce rich visualizations in dashboards that are automatically built based on
the knowledge we have put in place in the KG and an indicator ontology.
Both works make use of the proposed CCSV (Contextualized CSV) format
which we have designed to support not only raw les from data acquisition
instruments prior to be turned into knowledge in the KG, but also as an output
format for data in KGs. A CCSV le is a regular CSV le with a Turtle preamble
on top of it, which links the le contents (registers and columns) to a domain
ontology, thus preserving the semantics associated with the data. The CCSV
format has shown promising results as a way to bridge the gap between KGs
and data tools.
7
Approach
The main idea behind our approach is to provide a speci cation for the
construction of a scienti c KG along with processes in support of data analytics. The
key innovation and novel contribution is the ability to use knowledge in the KG
to provide data access and prepare datasets for data analytics, based on user
queries.
4 http://hadatac.org/ont/hasco#
In order to tackle Q1, we are gathering scienti c data analytics requirements
working in conjunction with data users in two domain areas: environmental and
urban. Based on the requirements, we are reusing and extending ontologies that
we believe will be capable of composing a base knowledge layer that can be
exploited by processes that aim to facilitate data analytics. Currently, we are using
HAScO as our base ontology for a scienti c KG as it makes use of PROV-O for
provenance tracking alongside VSTO-I [
12
] and proper extensions for registering
contextual knowledge. For Q2, in its turn, we are creating processes that are
able to retrieve desired data points by using the knowledge from our scienti c
HAScO-based KGs based on user queries. With that, we intend data users to
have direct access to data points instead of complete datasets in order to produce
more reliable data analytics.
Data in KGs are in triples format which is good for representing knowledge
but not so for data analytics tools which, most of the times, expect tabular data.
For Q3, we are working with two distinct approaches. First, we have discussed
the CCSV format in the previous section, which is able to be a way of serializing
data from KG together with its associated knowledge with the capability of
serving as an input format for intelligent applications that can take advantage of
that, as demonstrated by our preliminary results. We are continuously expanding
the format to handle new data analytics use-cases. Secondly, we are also studying
how to provide a programmatically way of accessing the desired data for analytics
from tools that support this feature.
8
Evaluation plan
We intend to validate H1 using state of the art KG evaluation approaches
discussed in [
20
]. For H2, we are gathering data analytics use cases and assessing
how the associated metadata facilitates the use of the data.
Ultimately, for H3 and H4, we intend to perform tests with data scientists
and eld specialists acting as users of our proposed KG and processes. Using
their data (preferably from di erent studies and sources), we intend to build
a scienti c KG adding the relevant metadata and then provide them tools for
querying the data and preparing datasets for their routine data analytics. Then,
questionnaires will be applied to measure how much our approach has eased
their tasks in contrast with their regular processes.
9
Re ections
To conclude, we have identi ed from the state of the art approaches that the
task of promoting data analytics from scienti c data in KGs is still in its early
stages. With this research, we intend to push this forward by proposing a KG
speci cation that not only is capable of tracking all the contextual knowledge
that is lost during data acquisition activities but is aligned with data analytics
requirements. More than that, we intend to exploit knowledge in KGs to be
able to return data points directly related to user queries instead of complete
datasets. Given this, we expect the outcome of this research to dramatically
decrease the data preparation e orts.
Acknowledgments
Advised by Prof. Vasco Furtado. Further thanks to Dr. Paulo Pinheiro and
Prof. Deborah L. McGuinness for the cooperation and invaluable feedback on
this work.
1.
Arenas
,
M.
,
CuencaGrau
,
B.
,
Kharlamov
,
E.
,
Marciuska
,
S.
,
Zheleznyakov
,
D.
:
Faceted search over RDF-based knowledge graphs
.
Web Semantics: Science, Services and Agents on the World Wide Web37
{
38
,
55
{74 (Mar
2016
)
2.
Ashburner
,
M.
,
Ball
,
C.A.
,
Blake
,
J.A.
,
Botstein
,
D.
,
Butler
,
H.
,
Cherry
,
J.M.
,
Davis
,
A.P.
,
Dolinski
,
K.
,
Dwight
,
S.S.
,
Eppig
,
J.T.
,
Harris
,
M.A.
,
Hill
,
D.P.
,
IsselTarver
,
L.
,
Kasarskis
,
A.
,
Lewis
,
S.
,
Matese
,
J.C.
,
Richardson
,
J.E.
,
Ringwald
,
M.
,
Rubin
,
G.M.
,
Sherlock
,
G.
:
Gene Ontology: tool for the uni cation of biology
.
Nature Genetics25
(
1
),
25
{29 (May
2000
)
3.
Auer
,
S.
,
Bizer
,
C.
,
Kobilarov
,
G.
,
Lehmann
,
J.
,
Cyganiak
,
R.
,
Ives
,
Z.
:
DBpedia: A Nucleus for a Web of Open Data
.
In: The Semantic Web
, pp.
722
{
735
. No. 4825
in Lecture Notes in Computer Science (Jan2007
)
4.
Beisswanger
,
E.
,
Schulz
,
S.
,
Stenzhorn
,
H.
,
Hahn
, U.:
BioTop: An upper domain ontology for the life sciences
.
Applied Ontology3
(
4
),
205
{212 (Jan
2008
)
5.
Bollacker
,
K.
,
Evans
,
C.
,
Paritosh
,
P.
,
Sturge
,
T.
,
Taylor
, J.:
Freebase: A Collaboratively Created Graph Database for Structuring Human Knowledge
.
In: Proceedings of the 2008 ACM SIGMOD International Conference on Management of Data
. pp.
1247
{
1250
(Jun
2008
)
6.
Callahan
,
A.
,
Cruz-Toledo
,
J.
,
Ansell
,
P.
,
Dumontier
,
M.:Bio2rdf Release 2: Improved Coverage, Interoperability and Provenance of Life Science Linked Data
.
In: The Semantic Web: Semantics and Big Data
. pp.
200
{
212
(May
2013
)
7.
Chen
,
Y.N.
,
Wang
,
W.Y.
,
Rudnicky
,
A.
:
Jointly Modeling Inter-Slot Relations by Random Walk on Knowledge Graphs for Unsupervised Spoken Language Understanding
.
In: Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies
. pp.
619
{
629
(Jun
2015
)
8.
Duan
,
W.
,
Chiang
,
Y.Y.
:
Building Knowledge Graph from Public Data for Predictive Analysis: A Case Study on Predicting Technology Future in Space and Time
.
In: Proceedings of the 5th ACM SIGSPATIAL International Workshop on Analytics for Big Geospatial Data
. pp.
7
{
13
(Oct
2016
)
9.
Ernst
,
P.
,
Siu
,
A.
,
Weikum
, G.:
KnowLife: a versatile approach for constructing a large knowledge graph for biomedical sciences
.
BMC Bioinformatics16
, 157 (May
2015
)
10.
Fox
,
M.S.:PolisGnosis Project: Representing and Analysing City Indicators
. Working Paper, Enterprise Integration Laboratory, University of Toronto (May
2015
), http://eil.utoronto.ca/wp-content/uploads/smartcities/papers/ PolisGnosis.pdf
11.
Fox
,
M.S.:The role of ontologies in publishing and analyzing city indicators
.
Computers, Environment and Urban Systems54
,
266
{279 (Nov
2015
)
12.
Fox
,
P.
,
McGuinness
,
D.L.
,
Cinquini
,
L.
,
West
,
P.
,
Garcia
,
J.
,
Benedict
,
J.L.
,
Middleton
,
D.
:
Ontology-supported scienti c data frameworks: The Virtual SolarTerrestrial Observatory experience
.
Computers & Geosciences35
(
4
),
724
{738 (Apr
2009
)
13.
Furtado
,
V.
,
Caminha
,
C.
,
Furtado
,
E.
,
Lopes
,
A.
,
Dantas
,
V.
,
Ponte
,
C.
,
Cavalcante
,
S.
:
Increasing the Likelihood of Finding Public Transport Riders that Face Problems Through a Data-Driven approach
.
arXiv:1705.03504 [cs] (Apr2017
)
14. ISO:
Sustainable development of communities { Indicators for city services and quality of life
.
ISO37120
:
2014
, International Organization for Standardization (May
2014
), http://www.iso.org/iso/catalogue_detail?csnumber=
62436
15. Je ery,
S.R.
,
Alonso
,
G.
,
Franklin
,
M.J.
,
Hong
,
W.
,
Widom
,
J.:A Pipelined Framework for Online Cleaning of Sensor Data Streams
.
In: 22nd International Conference on Data Engineering (ICDE'06)
. pp.
140
{
140
(Apr
2006
)
16.
Le-Phuoc
,
D.
, Nguyen Mau Quoc,
H.
,
NgoQuoc
,
H.
, Tran Nhat,
T.
,
Hauswirth
,
M.:The Graph of Things: A step towards the Live Knowledge Graph of connected things
.
Web Semantics: Science, Services and Agents on the World Wide Web37
{
38
,
25
{35 (Mar
2016
)
17.
Lopez
,
V.
,
Tommasi
,
P.
,
Kotoulas
,
S.
,
Wu
,
J.:QuerioDALI: Question Answering Over Dynamic and Linked Knowledge Graphs
.
In: The Semantic Web { ISWC 2016
. pp.
363
{
382
. Lecture Notes in Computer Science (Oct
2016
)
18.
Nickel
,
M.
,
Murphy
,
K.
,
Tresp
,
V.
,
Gabrilovich
, E.:
A Review of Relational Machine Learning for Knowledge Graphs
.
Proceedings of the IEEE104
(
1
),
11
{33 (Jan
2016
)
19.
Patil
,
D.J.
:
Data Jujitsu: The Art of Turning Data into Product
.
O'Reilly Media
,
1edn
.
(Nov2012
)
20.
Paulheim
, H.:
Knowledge graph re nement: A survey of approaches and evaluation methods
.
Semantic Web8
(
3
),
489
{508 (Jan
2017
)
21.
Pinheiro
,
P.
,
McGuinness
,
D.L.
,
Santos
, H.:
Human-Aware Sensor Network Ontology: Semantic Support for Empirical Data Collection
.
In: Proceedings of the 5th Workshop on Linked Science. Bethlehem
, PA, USA (Oct
2015
)
22.
Puiu
,
D.
,
Barnaghi
,
P.
, Tonjes, R., Kumper,
D.
,
Ali
,
M.I.
,
Mileo
,
A.
,
Parreira
,
J.X.
,
Fischer
,
M.
,
Kolozali
,
S.
,
Farajidavar
,
N.
,
Gao
,
F.
,
Iggena
,
T.
,
Pham
,
T.L.
,
Nechifor
,
C.S.
,
Puschmann
,
D.
,
Fernandes
,
J.:CityPulse: Large Scale Data Analytics Framework for Smart Cities
.
IEEE Access 4
,
1086
{
1108
(
2016
)
23.
Santos
,
H.
,
Dantas
,
V.
,
Furtado
,
V.
,
Pinheiro
,
P.
,
McGuinness
,
D.L.
:
From Data to City Indicators: A Knowledge Graph for Supporting Automatic Generation of Dashboards
.
In: The Semantic Web
. pp.
94
{
108
(May
2017
)
24.
Santos
,
H.
,
Furtado
,
V.
,
Pinheiro
,
P.
,
McGuinness
,
D.L.
:
Contextual Data Collection for Smart Cities
.
In: Proceedings of the Sixth Workshop on Semantics for Smarter Cities. Bethlehem
, PA, USA (Oct
2015
)
25.
Suchanek
,
F.M.
,
Kasneci
,
G.
,
Weikum
, G.:
YAGO: A Large Ontology from Wikipedia and WordNet
.
Web Semantics: Science, Services and Agents on the World Wide Web6
(
3
),
203
{217 (Sep
2008
)
26.
Vrandecic
,
D.
, Krotzsch, M.:
Wikidata
:
A FreeCollaborativeKnowledgebase
.
Commun. ACM57
(
10
),
78
{
85
(
2014
)