342
M. C. Ridley
manually entered or imported data about relations with customers and sales scripts.
Another classic example is a system for collecting sensor readings and storing them
in a transaction processing database along with OLAP cubes for analysts to prepare
reports for managers.
Even so, practical tasks often require direct application of entities, relations, entity
attributes, and rules that govern them. It is required in geospatial services, question answering, master data/record management systems, document management
systems, enterprise search, content management systems, and so on.
Online media monitoring is a classic case in different field of study. Since the
Internet is a kind of mirror for the real world, it contains the implicit mentions of
real-world events. Many related problems such as reconstitution and prediction of
events, explorative search, and early trend detection can successfully be formulated
in terms of ontology construction or ontology learning. Such ontology is a useful
source for computing domain-specific metrics like event importance, surges in topic
discussions, vocabulary diversity when discussing something, indirect links between
entities, semantical similarity of mentions in different languages.
In the present context, it is common to bring up the concept of a semantic Web
as a way of enriching data with meaning by means of a special markup and communicating metadata and knowledge among Web resources. If it was widely adopted,
semantic markup of Web pages would render information on the Internet machinereadable. In a sense, it helps to establish a connection between text and its conceived
meaning. One of the things that semantic Web brought to us is Web ontology language
(OWL), which is capable of representing classes, individuals, relations, and even
reasoning on top of described ontologies [1].
Similar ideas are found in the enterprise software domain. For example, an ordinary full-text search could be superseded by a semantic search capable of navigation,
filtering, grouping content with the same meaning, linking, and so on. Theoretical
foundation of such cases is extraction and representation of knowledge by means
of ontologies. These ontologies typically consist of entities or concepts, attributes,
relations, and slightly less common—axioms and rules.
Unfortunately, most ontologies are still constructed and even populated manually. This approach is expensive and time-consuming. It cannot be used at all in case
of large data inflow or when data has to be processed in a real-time fashion. Also,
manually constructed ontologies tend to be too general and thus form a habit of underestimating the power of applying ontologies to real-life tasks. It is also interesting that
early automatic ontology construction projects focused on generic ontologies as well:
usually, it was about extracting hyponyms/hypernyms and meronyms. Extraction of
non-taxonomic relations was not common for some time.
Automatic ontology construction is an important problem that requires extracting
the entities, concepts, and relations between occurring in a text corpus. For example,
analysis of corporate documents and e-mails leads to maintaining a knowledge
base with a map of all operations, document templates, workflows, etc. Analysis
of mentions in the Internet enables to build the predictive and descriptive models of
real-life events and their dynamics. By analyzing posts and chats of a person, one can
build the contextual profiles: what person talks about and how, what he or she likes,
M. C. Ridley
manually entered or imported data about relations with customers and sales scripts.
Another classic example is a system for collecting sensor readings and storing them
in a transaction processing database along with OLAP cubes for analysts to prepare
reports for managers.
Even so, practical tasks often require direct application of entities, relations, entity
attributes, and rules that govern them. It is required in geospatial services, question answering, master data/record management systems, document management
systems, enterprise search, content management systems, and so on.
Online media monitoring is a classic case in different field of study. Since the
Internet is a kind of mirror for the real world, it contains the implicit mentions of
real-world events. Many related problems such as reconstitution and prediction of
events, explorative search, and early trend detection can successfully be formulated
in terms of ontology construction or ontology learning. Such ontology is a useful
source for computing domain-specific metrics like event importance, surges in topic
discussions, vocabulary diversity when discussing something, indirect links between
entities, semantical similarity of mentions in different languages.
In the present context, it is common to bring up the concept of a semantic Web
as a way of enriching data with meaning by means of a special markup and communicating metadata and knowledge among Web resources. If it was widely adopted,
semantic markup of Web pages would render information on the Internet machinereadable. In a sense, it helps to establish a connection between text and its conceived
meaning. One of the things that semantic Web brought to us is Web ontology language
(OWL), which is capable of representing classes, individuals, relations, and even
reasoning on top of described ontologies [1].
Similar ideas are found in the enterprise software domain. For example, an ordinary full-text search could be superseded by a semantic search capable of navigation,
filtering, grouping content with the same meaning, linking, and so on. Theoretical
foundation of such cases is extraction and representation of knowledge by means
of ontologies. These ontologies typically consist of entities or concepts, attributes,
relations, and slightly less common—axioms and rules.
Unfortunately, most ontologies are still constructed and even populated manually. This approach is expensive and time-consuming. It cannot be used at all in case
of large data inflow or when data has to be processed in a real-time fashion. Also,
manually constructed ontologies tend to be too general and thus form a habit of underestimating the power of applying ontologies to real-life tasks. It is also interesting that
early automatic ontology construction projects focused on generic ontologies as well:
usually, it was about extracting hyponyms/hypernyms and meronyms. Extraction of
non-taxonomic relations was not common for some time.
Automatic ontology construction is an important problem that requires extracting
the entities, concepts, and relations between occurring in a text corpus. For example,
analysis of corporate documents and e-mails leads to maintaining a knowledge
base with a map of all operations, document templates, workflows, etc. Analysis
of mentions in the Internet enables to build the predictive and descriptive models of
real-life events and their dynamics. By analyzing posts and chats of a person, one can
build the contextual profiles: what person talks about and how, what he or she likes,