没有合适的资源?快使用搜索试试~ 我知道了~
Building Knowledge Graphs About Political Agents in the AgeofMis...
需积分: 11 2 下载量 92 浏览量
2019-08-09
15:01:24
上传
评论
收藏 1.38MB PDF 举报
温馨提示
Abstract. This paper presents the construction of a Knowledge Graph about relations between agents in a political system. It discusses the main modeling challenges, with emphasis on the issue of trust and provenance. Implementation decisions are also presented
资源推荐
资源详情
资源评论
Building Knowledge Graphs About Political
Agents in the Age of Misinformation
Daniel Schwabe, Carlos Laufer, Antonio Busson
Dept. of Informatics, PUC-Rio, R. M. de S. Vicente, 225, Rio de Janeiro, RJ 22453-900, Brazil
Abstract. This paper presents the construction of a Knowledge Graph about relations between agents in a political system. It
discusses the main modeling challenges, with emphasis on the issue of trust and provenance. Implementation decisions are also
presented
Keywords: Political Systems, trust, linked data, provenance, influence networks.
1. Introduction
Although the term “Knowledge Graph” (KG) was
introduced by Google in 2012
1
, graph-based databases
were available before this (e.g., Wordnet [28], DBPe-
dia [22], Yago [38], CYC [23], NELL [7], and addi-
tional ones (e.g. ConceptNet [37]) continue to be cre-
ated on a regular basis.
The majority of the largest published KGs are open-
ended, in the sense that they include facts in practically
any domain of knowledge. Consequently, for those
having an underlying semantic model (e.g. DBPedia,
Yago), their supporting ontologies are wide-ranging
and are constantly being updated to accommodate new
domains.
Another set of KGs are domain specific, and use
specialized ontologies to describe their data ( e.g. [36]
discusses several in the Life Sciences, [32] presents
ontologies for the music domain).
KGs differ also on the way they are built (popu-
lated). A few are curated (e.g., CYC), others rely on
crowdsourced information (e.g. Wikidata [39]), and
most extract information from structured, semi-struc-
tured or textual information harvested from the Web.
The multiplicity of sources and various extraction
approaches naturally raises the issue of data quality
and confronts the user of the data in the KG with the
1
https://googleblog.blogspot.com/2012/05/introducing-
knowledge graph-things-not.html
2
The expression “Se liga” in Portuguese has a colloquial mean-
ing of “be aware”, “pay attention to”, as well as “connect yourself”.
issue of trusting, or not, the information obtained from
the KG. For some types of information, for example in
case of online reviews online and social media, this
trust can have a direct effect on commercial success
(e.g. [4]). This highlights the fact that data, ultimately,
expresses a belief, opinion or point of view of some
agent.
This paper presents an approach to building a KG
in the domain of Political Agents, with a special em-
phasis on their different types of relations. This ap-
proach has been used in an initiative to build an open
KG about Political Agents in Brazil in the form of
Linked Data, named “Se Liga na Politica”
2
(SLNP).
The data in this KG is obtained from several sources,
in both automated and non-automated ways. Most of
the automated extraction is made from official sources,
such as the open data published by the House of Dep-
uties and by the Senate. In addition to such sources,
data may also be contributed by individuals, in
crowdsourced fashion.
1.1. Political Systems
Since very early times, men have organized them-
selves to form societies [34], naturally leading to the
formation of Political Systems, defined as “formal and
informal processes by which decisions are made
In Portuguese, it reads both as “pay attention to Politics” and “Con-
nect yourself to Politics”. A third (indirect) meaning is the reference
to Linked Data.
concerning the use, production and distribution of re-
sources in any given society”. Societies include
Agents – Persons and Organizations – that participate
in the processes of its Political System. In doing so,
they are driven, and constrained, by the various types
of relations that exist among them.
With the advent of the Information Society and the
Network Society [9] [13], accelerated by the wide-
spread adoption of the Internet and the Web, infor-
mation has become a vital resource, inextricably inter-
twined with the functioning of Political Systems.
Transparency, the quality that allows participants of
the society to know what are the particular processes
and agents that are being used in its functioning, is
generally regarded as a means to enable checks and
balances within Political Systems to prevent misuse by
any of the parties involved [19] [26]. One of the forms
to increase transparency within a Political System is to
provide information about its participants and their re-
lations, as a way to provide additional context when
analyzing their actions, and, ultimately, making deci-
sions.
In an ideal Linked Data World, a specific database
about political agents would not really be necessary,
as institutions responsible for each type of information
would publish them in Linked Data form, creating a
large KG. In practice, however, we are far from this –
the majority of the published information about Polit-
ical Systems is fragmented and incomplete. The focus
of the SLNP project is to establish the links between
the various “domains of knowledge” involved in de-
scribing Political Systems, taking care not to replicate
all of the information published by each source
(thereby, somehow, replacing it), focusing on charac-
terizing the relations between agents and often omit-
ting other properties that may be of interest of specific
communities.
1.2. Trust
Given the multiplicity of sources, and the nature of
the subject matter, this KG is designed so that facts are
seen as claims made by some agent, and therefore
provenance information becomes a “first class citizen”
of the domain. One of the main usages for this data-
base is to provide context information for news stories,
to allow readers to establish trust in the claimed facts
based on their own criteria.
The issue of trust has been prevalent in the Internet
since its popularization in the early 90s (see [16] for a
survey), with a focus on the lower layers of the
3
http://xmlns.com/foaf/spec/
Internet Architecture, emphasizing authentication.
More recently, with the advent of the Web and social
networks, the cybersphere, and society as a whole, has
become heavily influenced by information (and mis-
information) that flows in news sites and social net-
works in the Internet. There are many studies carried
out in several disciplines attempting to characterize
and understand the spread of information in the cy-
bersphere, and how this affects society (see [24] for an
overview). A more prominent aspect has been the
spread of “fake news”, actually a term used to refer to
several different misuses of information, as postulated
by Wardle in [39]. This has also been the focus of
much research and many initiatives (e.g. [10], [11],
[3]).
The original vision for the Semantic Web included
a “Trust” layer, although its emphasis was more on au-
thentication and validation, and static trust measures
for data. There have been many efforts in representing
trust, including computational models - a general sur-
vey can be found in [31]; [5] presents an excellent ear-
lier survey for the Semantic Web; and [35] surveys
trust in social networks. In the Linked Data world, it
is clear that facts in Semantic Web should be regarded
as claims rather than hard facts (e.g., [6]), which natu-
rally raises the issue of trust on those claims.
The remainder of this paper is organized as follows.
Section 2 presents the Domain Model for relations be-
teen agents in Political Systems; Section 3 presents a
model for the Trust Process supported by the SLNP
KG, and details how provenance information is repre-
sented and used; Section 4 briefly discusses imple-
mentation aspects, and Section 5 draws some conclu-
sions and points to ongoing and future work.
2. Domain Model
This section presents the POLARE ontologies that
characterize the relations between agents in Political
Systems. Before delving into details, some of the re-
quirements for the ontology and the rationale for the
design approach are discussed.
2.1. Methodological Approach
2.1.1. Ontologies vs vocabularies
As a general rule, preference was given to using
well-known ontologies, such as FOAF
3
, ORG
4
,
4
https://www.w3.org/TR/vocab-org/
SKOS
5
, Schema.org
6
, etc… as controlled vocabularies
to describe concepts in their respective domains. Pre-
cisely because these ontologies are very general, they
allow many possible uses within other ontologies.
POLARE, in many situations, defines specific ways
in which these vocabularies can be used for its pur-
poses; whenever the intended use was incompatible
with these ontologies, SLNP’s own vocabulary was
used. In addition, SLNP’s vocabulary also includes
terms to describe concepts not found in any of the bet-
ter-known controlled vocabularies.
POLARE is meant to be used to characterize data in
a Linked Data database. It is envisioned that this data
may be used in my different ways, for various pur-
poses. To allow such latitude, it was deliberately de-
signed in a “lightweight” fashion, with few specific in-
ference rules. It is understood that it is possible to ex-
tend it with a more “heavyweight” ontology by includ-
ing inference rules to further constrain the possible in-
terpretations, for use in specific situations.
One should also keep in mind that POLARE de-
scribes statements which are understood as claims be-
ing made by some agent (this is elaborated in section
3). Therefore, additional care must be taken when in-
cluding inference rules, as they may be expressing re-
strictions according do some particular point of view,
not necessarily accepted or agreed upon by all users.
The POLARE ontology includes several Datatype
properties, but since they are not so relevant for char-
acterizing relations, they will not be discussed in here.
2.1.2. OWL vs SKOS
OWL is a knowledge representation language, de-
signed to formulate, exchange and reason with
knowledge about a domain of interest. OWL can be
reasoned with by computer programs either to verify
the consistency of that knowledge or to make implicit
knowledge explicit [18].
An alternative approach to represent knowledge is
proposed by the SKOS ontology. A SKOS concept can
be viewed as an idea or notion; a unit of thought. How-
ever, what constitutes a unit of thought is subjective,
and this definition is meant to be suggestive, rather
than restrictive [27].
The approach used for SLNP is based on a hybrid
set of OWL vocabularies and SKOS Concept Schemes
to describe concepts in its domain. OWL is used to de-
fine more formal structures where the inference rules
can be used to make implicit knowledge. SKOS is
5
http://www.w3.org/TR/skos-reference
6
http://schema.org/docs/full.html
used to define concepts that are mainly used for re-
trieval and navigation tasks, and for which there are
many possible alternative schemes.
Several SKOS Concept Schemes have been identi-
fied that should complement the OWL classes defined
in the POLARE OWL vocabularies. They are used as
classifications for specific classes. The rationale for
choosing to use SKOS as opposed to OWL was based
on the generality vs specificity of the concept involved
– whenever the concept could be represented in many
different ways depending on the particular Political
System, SKOS was preferred. For example, the “clas-
sification” of an Organization can be made in many
different taxonomies, often non-mutually exclusive –
for instance, according to fiscal status, legal status,
type of ownership/control, etc… Such uses will be
highlighted throughout the description of POLARE.
Whenever possible, preference was given to utilizing
standard vocabularies.
2.2. Domain Models
2.2.1. People and Organizations
The central concepts in POLARE are Persons and
Organizations, as they are the Political Agents within
a Political System. Given the goal to characterize the
various kinds of relations between them, direct rela-
tions between Persons was the first focus, and then re-
lations between Persons and Organizations were ex-
amined, as they establish indirect relations between
Persons. The FOAF vocabulary was chosen to de-
scribe Persons, and the ORG vocabulary to describe
Organizations, adding relations in the POLARE ontol-
ogy as needed
7
.
The first kind of relations between persons are di-
rect family relations., which are modeled in POLARE
as Direct Relationships, shown in Error! Reference
source not found.. Rather than simply using an
owl:ObjectProperty, they are modeled via reification,
due to the need to qualify this relation with temporal
information. The directRelProp property allows spec-
ifying what is the family relation; its value, rather than
being an rdf:Property, is a skos:Concept, whose value
will be taken from a suitable Skos:ConceptScheme.
This allows inclusion of certain relations that may not
be “formally” accepted as a family relation but may be
of interest for some types of analyses, e.g., “co-habi-
tates”.
7
For readability purposes, we do not add a prefix to terms of PO-
LARE itself (e.g., pol:hasPost). Similarly, when it is clear which
ontology a term is from (e.g. foaf:Person), we omit the prefix.
剩余14页未读,继续阅读
资源评论
Jayxp
- 粉丝: 6
- 资源: 137
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功