A Case Study in Natural Language Based Web Search - Natural Language Processing and Text Mining - page 75

• Events involving a certain action type

“Action types” are groups of semantically linked actions. For example, query “ [Per-
son] > [Communication] > [Person] ” will retrieve all events involving communica-
tion between two people.
InFact's query syntax supports Boolean operators (i.e., AND, OR, NOT). For
example, the query:

Clinton NOT Hillary > visit OR travel to > [Location]

is likely to retrieve the travels of Bill Clinton , but not Hillary Clinton .
We can further constrain actions with modifiers, which can be explicit entities
or entity types, e.g., Paris or [location] . For example, the query

[Organization/Name] > buy > [Organization/Name]ˆ[money]

will only return results where a document mentions a specific monetary amount
along with a corporate acquisition. Similarly, the query

Bush <> meet<> Clinton ˆ[location]

will return results restricted to actions that occur in an explicit geographical loca-
tion.
We can also filter search results by specifying document-level constraints, in-
cluding:
• Document metadata tags - lists of returned actions, relationships or events are

restricted to documents that contain the specified metadata values.

• Boolean keyword expressions - lists of returned actions, relationships or events

are restricted to documents that contain the specified Boolean keyword expres-
sions.

For instance, a query like:

[Organization/Name] > buy > [Organization/Name]ˆ[money]; energy NOT oil

will return documents that mentions a corporate acquisition with a specific monetary
amount, and also contain the keyword “energy” but do not contain the keyword
“oil.”
InFact also provides a context operator for inter-clause linking. Suppose for
instance, that we want to retrieve all events where a plane crash kills a certain
number of passengers. The event could be spread over adjacent sentences, as in:
“The plane crashed shortly after take-off. As many as 224 people were killed.”
In this case, a query like:

* > kill > [numeric] ∼plane crash

will retrieve all plane crash events, regardless of whether they are contained in a
single or multiple, adjacent sentences.
InFact can also support synonyms and query expansion via custom ontologies.
In this case, InFact will automatically recognize the equivalence of entities or actions
that belong to the same ontology node.
The InFact Query Language rests on a flexible Java Search API. The Java
Search API allows us to programmatically concatenate search operators, package
and present them to the end user through a simpler interface.