23 Entity-Event Ontology Construction by Conceptualization …
343
what are links between topics according to this particular person, and so on. Using
such profiles, broad topics like “Politics” can be broken down into attitudes toward
specific politicians. Also, they can be employed to improve the personalization of
targeted ads.
Needless to say, same methods can be utilized in a broad spectrum of information
extraction tasks. For instance, it can be used to create domain-oriented search engines
and explorative search systems.
When information is extracted from documents according to some meta-ontology,
resulting database can then be used to provide rich navigation features, explicit and
implicit link analysis, grouping mentions by things they refer to, structured semantic
queries, temporal analysis, single-document and multi-document summarization,
and other tasks. The same can be said about some natural language understanding
tasks like question answering and textual entailment.
Our contributions are in the proposal of template-level and instance-level ontologies, introduction of entity-event ontologies, development of the method for construction entity-event ontologies from text corpora, implementing a full-featured system
based on the method and capable of analyzing mentions on the Internet, conducting
experiments for quality evaluation, and deployment of the system in civil aviation
organizations.
The chapter is organized as follows. Section 23.2 provides an overview of ontology
learning methods. Section 23.3 introduces a novel method of entity-event ontology
construction along with practical and theoretic considerations. Section 23.4 describes
system implementation and aspects of collecting data from the Internet. Section 23.5
concludes the chapter.
23.2 Related Work
Ontology construction, enrichment, and population attempts are all related to
ontology learning—the act of acquisition of a domain model from data [1]. Input
data can be in any form ranging from structured XML documents to semi-structured
HTML pages and unstructured raw natural language text. In the latter case, it is
ontology learning from text [2]. The task then is extracting conceptual knowledge
from text input and building or populating an ontology from it.
It is useful to note that ontologies can be very different: some of them are general,
some of them are domain-specific. Also, often they are not full-featured: for example,
restricting the original problem to a taxonomy case is rather popular. There is a
concept of a semantic spectrum that allows describing knowledge representations in
terms of expressiveness ranging from glossaries (simple lists of terms) to controlled
vocabularies, data dictionaries and thesauri, data models, taxonomies, and finally
full-features ontologies.
Ontology learning is a broad field of study that has different classifications [3, 4].
Usually, researchers divide methods based on the following characteristics:
343
what are links between topics according to this particular person, and so on. Using
such profiles, broad topics like “Politics” can be broken down into attitudes toward
specific politicians. Also, they can be employed to improve the personalization of
targeted ads.
Needless to say, same methods can be utilized in a broad spectrum of information
extraction tasks. For instance, it can be used to create domain-oriented search engines
and explorative search systems.
When information is extracted from documents according to some meta-ontology,
resulting database can then be used to provide rich navigation features, explicit and
implicit link analysis, grouping mentions by things they refer to, structured semantic
queries, temporal analysis, single-document and multi-document summarization,
and other tasks. The same can be said about some natural language understanding
tasks like question answering and textual entailment.
Our contributions are in the proposal of template-level and instance-level ontologies, introduction of entity-event ontologies, development of the method for construction entity-event ontologies from text corpora, implementing a full-featured system
based on the method and capable of analyzing mentions on the Internet, conducting
experiments for quality evaluation, and deployment of the system in civil aviation
organizations.
The chapter is organized as follows. Section 23.2 provides an overview of ontology
learning methods. Section 23.3 introduces a novel method of entity-event ontology
construction along with practical and theoretic considerations. Section 23.4 describes
system implementation and aspects of collecting data from the Internet. Section 23.5
concludes the chapter.
23.2 Related Work
Ontology construction, enrichment, and population attempts are all related to
ontology learning—the act of acquisition of a domain model from data [1]. Input
data can be in any form ranging from structured XML documents to semi-structured
HTML pages and unstructured raw natural language text. In the latter case, it is
ontology learning from text [2]. The task then is extracting conceptual knowledge
from text input and building or populating an ontology from it.
It is useful to note that ontologies can be very different: some of them are general,
some of them are domain-specific. Also, often they are not full-featured: for example,
restricting the original problem to a taxonomy case is rather popular. There is a
concept of a semantic spectrum that allows describing knowledge representations in
terms of expressiveness ranging from glossaries (simple lists of terms) to controlled
vocabularies, data dictionaries and thesauri, data models, taxonomies, and finally
full-features ontologies.
Ontology learning is a broad field of study that has different classifications [3, 4].
Usually, researchers divide methods based on the following characteristics: