346
M. C. Ridley
used by employees to navigate through complex procedures, sets of legal templates,
contract provisions, and so on.
Speaking of news, politics, general or corporate events, legal announcements, and
social media posts, on some level one can say that they share a lot of commonalities.
One way of capturing these commonalities is a method of entity-event ontologies.
The method of entity-event ontologies is based on the idea that as something
important happens in the real world, it leaves traces in multiple mentions in documents, news, social network posts, and so on. Real-world events are scattered across
numerous mentions of them. All those mentions combined describe a single reallife event—for example, company acquisition, product announcement, legal actions,
protests and attacks, or natural disasters.
An example of suitable ontology for describing real-life events is rich event
ontology [6]. Unlike general ontologies, it does not underrepresent events and
treats them explicitly by introducing concepts like participants, causal and temporal
relations, special relations like HAS-PRECONDITION, HAS-RESULT, HASSUBEVENT, and so on.
Similar ideas are also implemented in Thomson Reuters (Refinitiv) knowledge
graph for financial and capital markets analysis. It models many real-life event
and entity types like organizations, quotes, regulators, assets, supply chains, deals,
industries, exchanges, company officers, and so on.
In our system, we used entity-event ontology that is based on real-life events and
entities, their mentions (instances) in texts, and context information derived from the
text and other sources. Each real-life event instance has some attributes depending on
its type. For example, natural disaster events have attributes disaster type, location,
time, deaths, other casualties, related entities, and mentions.
During pilot operation of our system, we have identified features not universally
present in similar systems, but useful and important for solving practical knowledge
base-related tasks:
• Spatio-temporal analysis can be utilized for creating chronologies of how events
unfold and how they were discussed. Also, it helps to identify and track planned
events like protests. Both time and place can be specified on their own or be
relative to other temporal or spatial statements.
• Time and geographical “intervals” and hierarchy. When considering all elections
in Russia, it is important to consider mention in all of its cities and regions.
Also, time intervals can have different precision ranging from seconds to years
and centuries. Temporal and spatial hierarchies enhance quality of merging
different mentions of presumably same real-life events when they have different
descriptions.
• Language-independent ontology. Although text processing methods in the system
are language-dependent, the resulting facts are independent, so that one real-life
event can have multiple mentions in different languages.
• Special events for indirect speech and quotations are handy for analysts as they
get an option to compare different judgments on the same topic, validate expert
opinions, and search for discrepancies in someone’s opinion. As they also have
M. C. Ridley
used by employees to navigate through complex procedures, sets of legal templates,
contract provisions, and so on.
Speaking of news, politics, general or corporate events, legal announcements, and
social media posts, on some level one can say that they share a lot of commonalities.
One way of capturing these commonalities is a method of entity-event ontologies.
The method of entity-event ontologies is based on the idea that as something
important happens in the real world, it leaves traces in multiple mentions in documents, news, social network posts, and so on. Real-world events are scattered across
numerous mentions of them. All those mentions combined describe a single reallife event—for example, company acquisition, product announcement, legal actions,
protests and attacks, or natural disasters.
An example of suitable ontology for describing real-life events is rich event
ontology [6]. Unlike general ontologies, it does not underrepresent events and
treats them explicitly by introducing concepts like participants, causal and temporal
relations, special relations like HAS-PRECONDITION, HAS-RESULT, HASSUBEVENT, and so on.
Similar ideas are also implemented in Thomson Reuters (Refinitiv) knowledge
graph for financial and capital markets analysis. It models many real-life event
and entity types like organizations, quotes, regulators, assets, supply chains, deals,
industries, exchanges, company officers, and so on.
In our system, we used entity-event ontology that is based on real-life events and
entities, their mentions (instances) in texts, and context information derived from the
text and other sources. Each real-life event instance has some attributes depending on
its type. For example, natural disaster events have attributes disaster type, location,
time, deaths, other casualties, related entities, and mentions.
During pilot operation of our system, we have identified features not universally
present in similar systems, but useful and important for solving practical knowledge
base-related tasks:
• Spatio-temporal analysis can be utilized for creating chronologies of how events
unfold and how they were discussed. Also, it helps to identify and track planned
events like protests. Both time and place can be specified on their own or be
relative to other temporal or spatial statements.
• Time and geographical “intervals” and hierarchy. When considering all elections
in Russia, it is important to consider mention in all of its cities and regions.
Also, time intervals can have different precision ranging from seconds to years
and centuries. Temporal and spatial hierarchies enhance quality of merging
different mentions of presumably same real-life events when they have different
descriptions.
• Language-independent ontology. Although text processing methods in the system
are language-dependent, the resulting facts are independent, so that one real-life
event can have multiple mentions in different languages.
• Special events for indirect speech and quotations are handy for analysts as they
get an option to compare different judgments on the same topic, validate expert
opinions, and search for discrepancies in someone’s opinion. As they also have